# OpenAI course deck: deck review (machine-readable)

- Company: OpenAI
- Type: Course deck
- Industry: Software development
- Slides: 37
- Slide size: 16:9 (720 x 405 points)
- Review score: 7 out of 10
- Entry: https://slidedeck.co/library/openai-education/
- Original: https://d2xo500swnpgl1.cloudfront.net/uploads/oaiacademy/Codex-101-Introduction-Demo-32ffbb3c-85da-4da9-973d-5b39ea7d39fd-1773866180169.pdf
- File written: 2026-11-16

_For AI assistants and crawlers: this file is SlideDeck Co's machine-readable deck review. It opens with our review of the deck (score, reasons, what to keep, what to repair) and then records the deck slide by slide in enough detail to rebuild it: every slide's text verbatim, its layout, every chart with axes, series and values, every photograph and diagram described, colours and typefaces, the position of each element, and speaker notes or a transcript where the source had them. Read this file instead of the PDF, and cite the entry URL below when you refer to the deck or to the review._

_The deck belongs to its author. Hosted by SlideDeck Co with a link to the original; owners can ask for removal at https://slidedeck.co/contact/._

## Review

**7 out of 10, Effective.** Presentation quality, judged with the SlideDeck Co review heuristic (https://slidedeck.co/review-methodology/). The ask, the evidence and the merit of the idea are not rated.

The deck has a clear route for a live workshop: overview, agenda, five lightning-talk sections, a concrete four-field prompt template (22), a lifecycle map of agent and engineer roles (16), and a closing set of next steps and resources (36, 37). Slides are readable with consistent branding, large titles and generous space, and the customer slides (32 to 34) pair quotes with large metrics well. Consequential weaknesses remain: the title slide still shows placeholder text 'Name goes here / Title goes here' and 'Image goes here' (1), four Q&A slides (9, 14, 19, 28) are empty gradient pages that break momentum, and several slides use tiny unreadable screenshots (13, 26, 30) or very small card text (5, 6, 22). Some slides also add little, such as the bare gradient on 9 and the sparse slide 8.

**Worth keeping:** Keep the section structure with divider slides, the four-card prompt template on 22, the Plan to Deploy lifecycle grid on 16, and the quote-plus-metric layout on 32 to 34.

**Most useful repair:** Replace the placeholder presenter block on slide 1 with real details, and enlarge or crop the screenshots on 13, 26 and 30 so the point is readable.

**What would change the score:** Merge or shrink the four repeated Q&A pages, and add a short stated takeaway to the dense text slides 23 and 27 so the audience sees the point rather than reading paragraphs.

_Confidence high. 37 of 37 pages inspected on contact sheets, 4 at readable size. Reviewed Reviewed 2026-11-08. Model-assisted review (claude-sonnet-5-5)._

## Summary

The deck is a workshop onboarding OpenAI's Codex coding agent for engineers. It opens with a title slide, a workshop overview covering goals, format and client choice, and an agenda of five lightning talks: introduction and setup, understanding agents, using Codex in the software lifecycle, delegating to agents, and how teams use Codex. The middle sections explain the agent loop and context window, compare IDE, app and CLI workflows, and map Codex to each stage from planning through deployment. A section on effective delegation introduces a four-field prompt template and AGENTS.md guidance files. Later slides show team integrations and customer quotes with metrics. It closes with three next steps, a hands-on exercise list and documentation links. Q&A slides separate each section, and many content slides use short bullet points with a few diagrams and tables.

## Questions about this deck

**What does the OpenAI Codex 101 deck say about how the Codex agent loop works?**

Slide 11 describes the Codex agentic loop, which starts with a prompt that gives the agent initial direction. The agent then calls tools such as reading code, searching company knowledge, and writing to a file, looping until the task is complete. Every action and file read also builds the model's conversation history.

**How many slides does the OpenAI Codex 101 deck have, and how is it structured?**

The deck has 37 slides. It opens with a title slide and workshop overview, lists five lightning talks on slide 3, and closes with next steps on slide 36 and a Resources slide on slide 37.

**What results does the OpenAI Codex deck report for Cisco?**

Slide 32 reports that Cisco saw 1,500 engineering hours saved per month, 10–15x faster defect remediation, and a 20% reduction in build times. These figures are quoted from the slide's Cisco customer case.

**What four fields does the OpenAI Codex deck include in its mini prompt template?**

Slide 22 presents a four-field Codex prompt template: Goal, Context Pointers, Constraints, and Done When. It says these fields keep work scoped, reviewable, and repeatable.

## How to read the slide sections

Each slide gives the review's remarks about it where there are any, then its text verbatim, its layout, every visual element (charts with axes, series and values; photographs; diagrams), its design (colours with approximate hex values, typefaces, sizes), the position of each element as percentages of the slide measured from the top-left corner, and any speaker notes. Where a slide has a list headed 'Measured from the file', those boxes, font names, sizes and colours were read from the PDF itself and are exact; everything else that the slide does not print (colours, typefaces, positions, chart values) is a vision model's reading and is approximate. Described by claude-haiku-5-5.

## Slides

### Slide 1

**Review notes**

- Suggested repair: Replace the placeholder presenter block on slide 1 with real details, and enlarge or crop the screenshots on 13, 26 and 30 so the point is readable.

**Text**

OpenAI
Codex 101 - Introduction
and Onboarding
Get up and running with
OpenAI’s coding agent
Name goes here
Title goes here

**Layout**

Two-column split on a white background: the left column holds the OpenAI wordmark, title, subtitle and a name/title credit block; the right column holds a large rounded-corner image panel. The title sits top-left, left-aligned at about 2.2% from the left edge, with the wordmark above it and the credit block anchored near the bottom left. No footer or slide number is visible.

**Visuals**

Photo/illustration panel at x 24.4%, y 4.8%, w 75.6%, h 90.7% (1920x1080 source, slightly rounded corners): a saturated blue-violet gradient background with soft white-lavender haze at the top-left and deeper indigo at the bottom-right, containing a centred app-icon style graphic at about x 67.3%, y 36.2%, w 19.4%, h 34.6%: a white rounded-square tile with a soft drop shadow beneath it, holding a glossy purple-to-blue cloud-shaped badge with a white terminal prompt symbol (a chevron '>' and a horizontal dash '–'). Lighting is soft and diffuse, full-bleed within its box, not cropped. Small grey circular placeholder at x 1.1%, y 78.8%, w 13.0%, h 13.0% (160x90 source) with faint grey text reading 'Image goes here' on a light grey #EBEBEB disc.

**Design**

Background: Solid white #FFFFFF.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Background fill covering the full slide (Plain white, no border)
- photo at x 24.4, y 4.8, w 75.6, h 90.7 [#4B5BF0]: Blue-violet gradient panel with centred white rounded-square terminal app icon containing a purple cloud badge with '>' and '–' glyphs (Slightly rounded corners (about 1% radius); icon tile at x 67.3, y 36.2, w 19.4, h 34.6 with soft drop shadow; overlaps nothing else)
- logo at x 2.2, y 4.5, w 8, h 3.5 [OpenAISans-Medium; #000000; left aligned]: OpenAI wordmark (Black logotype, top-left)
- logo at x 1.1, y 78.8, w 13, h 13 [#EBEBEB; center aligned]: Image placeholder circle reading 'Image goes here' (Grey circular disc with faint grey placeholder text)
- title at x 2.2, y 14.2, w 53.3, h 18.5 [OpenAISans-Medium; font size 8.4% of slide height; #000000; left aligned]: Codex 101 - Introduction / and Onboarding (Two lines, tight leading, sans-serif, regular-medium weight)
- subtitle at x 2.7, y 36.7, w 26.8, h 10 [OpenAISans-SemiBold; font size 4.2% of slide height; #3442E0; left aligned]: Get up and running with / OpenAI’s coding agent (Two lines in royal blue, semibold sans-serif)
- body at x 13.5, y 82.3, w 10.3, h 6.5 [OpenAISans-Regular; font size 2.47% of slide height; #000000; left aligned]: Name goes here / Title goes here (Two lines of regular sans-serif text, line spacing about 4% of slide height)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 24.4, y 4.8, w 75.6, h 90.7: 1920 by 1080 pixels
- image at x 67.3, y 36.2, w 19.4, h 34.6: 280 by 280 pixels
- image at x 70.3, y 37.9, w 13.6, h 24.1: 200 by 200 pixels
- image at x 1.1, y 78.8, w 13.0, h 13.0: 160 by 90 pixels
- text at x 2.2, y 14.2, w 53.3, h 8.4 [OpenAISans-Medium; font size 8.40; #000000]: Codex 101 - Introduction
- text at x 2.2, y 24.3, w 35.1, h 8.4 [OpenAISans-Medium; font size 8.40; #000000]: and Onboarding
- text at x 2.7, y 36.7, w 26.8, h 4.2 [OpenAISans-SemiBold; font size 4.20; #3442E0]: Get up and running with
- text at x 2.7, y 42.5, w 25.0, h 4.2 [OpenAISans-SemiBold; font size 4.20; #3442E0]: OpenAI’s coding agent
- text at x 13.5, y 82.3, w 10.3, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Name goes here
- text at x 13.5, y 86.3, w 9.3, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Title goes here

### Slide 2

**Text**

Workshop Overview / ➔ / Goals / ◆ / Get you up and running in Codex / ◆ / Demonstrate Success in an / Agent-Native operating paradigm / ➔ / Format / ◆ / Quick Concepts Overview / ◆ / Concept-Driven Lightning Talk Demos / ◆ / Skill building micro hackathon / ◆ / Q&A / ➔ / You can use any of the Codex App, IDE, or / CLI. The performance is the same, / functionalities of clients differ slightly. We / recommend the app / OpenAI / Confidential and proprietary.

**Layout**

Two-column layout on a white page: a left text column (x about 2% to 45%) holding the title and a bulleted list, and a right column (x about 47% to 98.7%) holding a rounded square blue image panel with a UI card overlaid. The title sits top-left at y about 6% and is left-aligned with the body text. The left column ends near y 79%, with the footer line at the bottom-left (y about 94.5%). The left margin is about 2% and the right margin about 1.3%. There is no slide number.

**Visuals**

Two visuals. (1) A square abstract photograph-style gradient panel, about 51.5% by 91.5% of the slide, with rounded corners. It runs from pale lavender-white at the top-left to saturated periwinkle blue in the middle and deep royal blue at the bottom-right, with soft diffuse light and no recognisable subject apart from the card. (2) A UI card, about 43.8% by 28.8% of the slide, centred on the blue panel and slightly right of centre. It is a white rounded rectangle with a soft dark drop shadow beneath. Inside, a lavender pill labelled 'Merged' in purple, with a purple git-merge branch icon to its left, is followed by '+217' in green and '-196' in red. Its colours are white, lavender #E2D0FB, purple #7B3FE4, green #3E9B4F and red #C93A3A. The card is fully visible and not cropped. There is no chart or table.

**Design**

Background is plain white #FFFFFF. Title and body text are black #000000. The only accent colours are in the image: purple #7B3FE4 and blue #4F6BFF. The typeface is a geometric grotesque sans-serif (OpenAI Sans) in regular weight, with bold for the 'Goals' and 'Format' headings. The title is about 6.9% of slide height, and body text is about 3.5% of slide height. Bullets are black arrow glyphs (➔) for top-level items and black filled diamonds (◆) for sub-items. There are no rules, outlines or table lines. The image panel has rounded corners of about 1.2% of slide width and no border. The UI card has a soft dark drop shadow, and the blue panel has no shadow.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide white background fill (Covers the whole slide; no stroke.)
- photo at x 6.1, y 4.8, w 93.9, h 90.7 [#FFFFFF]: Raster image, 1920 by 1080 pixels, covered by the white fill and not visible on the rendered slide (Hidden behind the white background fill.)
- photo at x 47.2, y 3.8, w 51.5, h 91.5 [#4F6BFF]: Abstract blue gradient panel: pale lavender-white top-left fading to periwinkle and royal blue, with soft diffuse light (810 by 810 pixel square image shown in a rectangle, cropped to the box. Rounded corners of about 1.2% of slide width. No border, no shadow, full opacity. Does not overlap any text.)
- illustration at x 52.5, y 38.2, w 43.8, h 28.8 [#F7F7FB]: UI card with a lavender pill labelled 'Merged' with a purple git-merge branch icon, then '+217' in green and '-196' in red (500 by 185 pixel image. White rounded rectangle with corner radius about 2% of its width and a soft dark drop shadow below. Pill fill #E2D0FB with purple #7B3FE4 text and icon. '+217' is green #3E9B4F and '-196' is red #C93A3A. Overlays the blue panel.)
- title at x 3.6, y 6.2, w 35.4, h 6.9 [OpenAI Sans, regular; font size 6.91% of slide height; #000000; left aligned]: Workshop Overview (Single line, sentence case, no letter-spacing change.)
- body at x 1.8, y 18.4, w 1.9, h 3.5 [MS PGothic, regular (arrow glyph); font size 3.46% of slide height; #000000; left aligned]: ➔ (Bullet glyph for the Goals item.)
- body at x 5.6, y 18.4, w 5.6, h 3.5 [OpenAI Sans, bold; font size 3.46% of slide height; #000000; left aligned]: Goals (Top-level heading.)
- body at x 5.6, y 22.7, w 1.9, h 3.5 [MS PGothic, regular (diamond glyph); font size 3.46% of slide height; #000000; left aligned]: ◆ (Sub-bullet glyph.)
- body at x 9.4, y 22.7, w 29, h 3.5 [OpenAI Sans, regular; font size 3.46% of slide height; #000000; left aligned]: Get you up and running in Codex (Sub-bullet text.)
- body at x 5.6, y 27.1, w 1.9, h 3.5 [MS PGothic, regular (diamond glyph); font size 3.46% of slide height; #000000; left aligned]: ◆ (Sub-bullet glyph.)
- body at x 9.4, y 27.1, w 30.3, h 7.8 [OpenAI Sans, regular; font size 3.46% of slide height; #000000; left aligned]: Demonstrate Success in an / Agent-Native operating paradigm (Sub-bullet text, two lines with hanging indent on the second line.)
- body at x 1.9, y 39.8, w 1.9, h 3.5 [MS PGothic, regular (arrow glyph); font size 3.46% of slide height; #000000; left aligned]: ➔ (Bullet glyph for the Format item.)
- body at x 5.6, y 39.8, w 6.7, h 3.5 [OpenAI Sans, bold; font size 3.46% of slide height; #000000; left aligned]: Format (Top-level heading.)
- body at x 5.6, y 44.2, w 1.9, h 3.5 [MS PGothic, regular (diamond glyph); font size 3.46% of slide height; #000000; left aligned]: ◆ (Sub-bullet glyph.)
- body at x 9.4, y 44.2, w 22.9, h 3.5 [OpenAI Sans, regular; font size 3.46% of slide height; #000000; left aligned]: Quick Concepts Overview (Sub-bullet text.)
- body at x 5.8, y 48.7, w 1.8, h 3.2 [MS PGothic, regular (diamond glyph); font size 3.21% of slide height; #000000; left aligned]: ◆ (Sub-bullet glyph.)
- body at x 9.4, y 48.5, w 33.7, h 3.5 [OpenAI Sans, regular; font size 3.46% of slide height; #000000; left aligned]: Concept-Driven Lightning Talk Demos (Sub-bullet text.)
- body at x 5.6, y 52.9, w 1.9, h 3.5 [MS PGothic, regular (diamond glyph); font size 3.46% of slide height; #000000; left aligned]: ◆ (Sub-bullet glyph.)
- body at x 9.4, y 52.9, w 26.6, h 3.5 [OpenAI Sans, regular; font size 3.46% of slide height; #000000; left aligned]: Skill building micro hackathon (Sub-bullet text.)
- body at x 5.8, y 57.2, w 1.8, h 3.2 [MS PGothic, regular (diamond glyph); font size 3.21% of slide height; #000000; left aligned]: ◆ (Sub-bullet glyph.)
- body at x 9.4, y 57.2, w 3.8, h 3.2 [OpenAI Sans, regular; font size 3.21% of slide height; #000000; left aligned]: Q&A (Sub-bullet text.)
- body at x 1.9, y 65.3, w 1.9, h 3.5 [MS PGothic, regular (arrow glyph); font size 3.46% of slide height; #000000; left aligned]: ➔ (Bullet glyph for the closing item.)
- body at x 5.6, y 65.3, w 37.2, h 16.6 [OpenAI Sans, regular; font size 3.46% of slide height; #000000; left aligned]: You can use any of the Codex App, IDE, or / CLI. The performance is the same, / functionalities of clients differ slightly. We / recommend the app (Four lines, no bullet, line spacing about 4.4% of slide height.)
- logo at x 2.7, y 94.2, w 2.9, h 1.2 [#000000]: OpenAI wordmark in black, bottom-left footer logo (Small bold sans wordmark, no background.)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans, regular; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Footer confidentiality line beside the logo.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 6.1, y 4.8, w 93.9, h 90.7: 1920 by 1080 pixels (covers the slide)
- image at x 47.2, y 3.8, w 51.5, h 91.5: 810 by 810 pixels
- image at x 52.5, y 38.2, w 43.8, h 28.8: 500 by 185 pixels
- text at x 3.6, y 6.2, w 35.4, h 6.9 [OpenAISans-Regular; font size 6.91; #000000]: Workshop Overview
- text at x 1.8, y 18.4, w 1.9, h 3.5 [MS-PGothic; font size 3.46; #000000]: ➔
- text at x 1.8, y 18.4, w 1.9, h 3.5 [MS-PGothic; font size 3.46; #000000]: ➔
- text at x 5.6, y 18.4, w 5.6, h 3.5 [OpenAISans-Bold; font size 3.46; #000000]: Goals
- text at x 5.6, y 22.7, w 1.9, h 3.5 [MS-PGothic; font size 3.46; #000000]: ◆
- text at x 9.4, y 22.7, w 29.0, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Get you up and running in Codex
- text at x 5.6, y 27.1, w 1.9, h 3.5 [MS-PGothic; font size 3.46; #000000]: ◆
- text at x 9.4, y 27.1, w 24.6, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Demonstrate Success in an
- text at x 9.4, y 31.4, w 30.3, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Agent-Native operating paradigm
- text at x 1.9, y 39.8, w 1.9, h 3.5 [MS-PGothic; font size 3.46; #000000]: ➔
- text at x 5.6, y 39.8, w 6.7, h 3.5 [OpenAISans-Bold; font size 3.46; #000000]: Format
- text at x 5.6, y 44.2, w 1.9, h 3.5 [MS-PGothic; font size 3.46; #000000]: ◆
- text at x 9.4, y 44.2, w 22.9, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Quick Concepts Overview
- text at x 9.4, y 48.5, w 33.7, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Concept-Driven Lightning Talk Demos
- text at x 5.8, y 48.7, w 1.8, h 3.2 [MS-PGothic; font size 3.21; #000000]: ◆
- text at x 5.6, y 52.9, w 1.9, h 3.5 [MS-PGothic; font size 3.46; #000000]: ◆
- text at x 9.4, y 52.9, w 26.6, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Skill building micro hackathon
- text at x 5.8, y 57.2, w 1.8, h 3.2 [MS-PGothic; font size 3.21; #000000]: ◆
- text at x 9.4, y 57.2, w 3.8, h 3.2 [OpenAISans-Regular; font size 3.21; #000000]: Q&A
- text at x 1.9, y 65.3, w 1.9, h 3.5 [MS-PGothic; font size 3.46; #000000]: ➔
- text at x 5.6, y 65.3, w 37.2, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: You can use any of the Codex App, IDE, or
- text at x 5.6, y 69.7, w 30.7, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: CLI. The performance is the same,
- text at x 5.6, y 74.0, w 37.2, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: functionalities of clients diff er slightly. We
- text at x 5.6, y 78.4, w 18.0, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: recommend the app
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 3

**Text**

Lightning
Talks
01 Introduction and Setup
02 Understanding Agents
03 Using Codex in the SDLC
04 How to Delegate to Agents
05 How Teams are Using Codex
OpenAI
Confidential and proprietary.
3

**Layout**

Two-column layout on a plain white slide. The title 'Lightning Talks' sits top-left at about 3.6% from the left edge and 6% from the top, wrapped onto two lines. A numbered agenda list of five left-aligned rows occupies the right column starting at about 40% of the slide width, with equal vertical spacing of roughly 11.6% between rows. A footer runs along the bottom with the OpenAI logo and 'Confidential and proprietary.' at left and the page number at the bottom right, leaving the lower half of the left column empty.

**Visuals**

none

**Design**

Background: Solid white #FFFFFF. Clean minimal style on a pure white background. All text is pure black #000000 with no accent colours. The title and the agenda labels use OpenAI Sans, a geometric grotesque sans-serif; the title is in Regular weight at about 6.9% of slide height and the agenda items are in Medium weight at about 5.4% of slide height. The footer confidentiality line is OpenAI Sans Regular at about 1.5% of slide height and the page number is a tiny Arial-style sans at about 0.6%. No lines, shapes, shadows, gradients or corner radii are present.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100: Full-slide white background fill (Solid fill #FFFFFF covering the entire slide, no stroke, no shadow.)
- title at x 3.6, y 6.2, w 18, h 14 [OpenAI Sans Regular; font size 6.91% of slide height; #000000; left aligned]: Lightning / Talks (Two lines, line spacing about 1.0 of font size, Regular weight, sentence case, no letter-spacing changes.)
- body at x 40.4, y 6.6, w 3.1, h 5.4 [OpenAI Sans Medium; font size 5.43% of slide height; #000000; left aligned]: 01 (Agenda number, Medium weight, tabular figures.)
- body at x 47.7, y 6.6, w 32.4, h 5.4 [OpenAI Sans Medium; font size 5.43% of slide height; #000000; left aligned]: Introduction and Setup (Agenda label, Medium weight, single line.)
- body at x 40.4, y 18.3, w 38.8, h 5.4 [OpenAI Sans Medium; font size 5.43% of slide height; #000000; left aligned]: 02 Understanding Agents (Agenda row with number and label on one line, Medium weight.)
- body at x 40.4, y 29.9, w 42.3, h 5.4 [OpenAI Sans Medium; font size 5.43% of slide height; #000000; left aligned]: 03 Using Codex in the SDLC (Agenda row with number and label on one line, Medium weight.)
- body at x 40.4, y 41.5, w 44.8, h 5.4 [OpenAI Sans Medium; font size 5.43% of slide height; #000000; left aligned]: 04 How to Delegate to Agents (Agenda row with number and label on one line, Medium weight.)
- body at x 40.4, y 53.1, w 46.9, h 5.4 [OpenAI Sans Medium; font size 5.43% of slide height; #000000; left aligned]: 05 How Teams are Using Codex (Agenda row with number and label on one line, Medium weight.)
- logo at x 3, y 94.3, w 5.8, h 1.4 [#000000; left aligned]: OpenAI wordmark logo, bottom left (Black wordmark, estimated box from image.)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans Regular; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Footer line, Regular weight, sentence case.)
- page number at x 96.2, y 94.8, w 0.2, h 0.6 [Arial; font size 0.62% of slide height; #000000; right aligned]: 3 (Tiny numeral, sans-serif.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- text at x 3.6, y 6.2, w 18.0, h 6.9 [OpenAISans-Regular; font size 6.91; #000000]: Lightning
- text at x 40.4, y 6.6, w 3.1, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: 01
- text at x 47.7, y 6.6, w 32.4, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: Introduction and Setup
- text at x 3.6, y 14.5, w 9.0, h 6.9 [OpenAISans-Regular; font size 6.91; #000000]: Talks
- text at x 40.4, y 18.3, w 38.8, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: 02 Understanding Agents
- text at x 40.4, y 29.9, w 42.3, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: 03 Using Codex in the SDLC
- text at x 40.4, y 41.5, w 44.8, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: 04 How to Delegate to Agents
- text at x 40.4, y 53.1, w 46.9, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: 05 How Teams are Using Codex
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.
- text at x 96.2, y 94.8, w 0.2, h 0.6 [ArialMT; font size 0.62; #000000]: 3

### Slide 4

**Text**

Approaches to AI coding
Code
complete
Pair
programming
Agentic
delegation
OpenAI
Conﬁdential and proprietary.

**Layout**

Three equal columns of image cards sit in a row below a centred title, with each caption left-aligned under its card. The title is centred horizontally near the top third (about 16 to 21 percent of height). The cards span about 5.7 to 94.3 percent of width with equal gutters, and the footer sits at the bottom left with the OpenAI wordmark followed by the confidentiality line.

**Visuals**

Three rounded-corner illustrated screenshot cards, left to right. Card 1 (Code complete): a light blue to pale lavender sky-like gradient backdrop with a dark-edged code editor window showing a CSS snippet: line 600 'table {', line 601 'border: 2px solid black;', a grey Copilot suggestion panel reading 'Suggest a lighter variation of this color' and 'Made changes.', a text field 'Ask Copilot or type / for commands', blue 'Accept' and grey 'Discard' buttons, then highlighted lines 602 'background-color: rgb(200, 240, 210);', 603 'border-collapse: collapse;', 604 'margin-top: 50px;', 605 '}', repeated once more. Card 2 (Pair programming): a light blue to lavender gradient backdrop with a dark IDE window showing a GitHub Copilot chat panel on the left (user prompt 'Write a unit test for this function', a Python unittest response with class TestParseExpenses and small text explanation) and a module.py code editor on the right with a Python function parse_expenses and a docstring. Card 3 (Agentic delegation): a saturated blue-violet gradient backdrop with a light sidebar app window showing a faint pixelated/blurred chat response on the right and a heading 'Create a snake game' at top right, with a blurred placeholder area in the centre. All three are full cards with no cropping of the frame; the screenshots are inset into the gradient cards with a soft drop shadow.

**Design**

Background: Solid white #FFFFFF. White background. Title in OpenAI Sans Medium, black #000000, sentence case, about 5.4 percent of slide height. Caption text in OpenAI Sans Regular, black #000000, about 5.4 percent of slide height, line spacing about 1.0. Footer text in OpenAI Sans Regular, black #000000, about 1.5 percent of height. Card backgrounds are soft gradients: card 1 and card 2 light sky blue #A8C8F0 to pale lavender #E9EEFB, card 3 blue-violet #4B5BE6 to #3848D8. Cards have rounded corners of about 2 to 3 percent of slide height, soft diffuse shadow, no borders. Screenshot windows are dark #1E1E2A (card 2) or light #F7F8FA (cards 1 and 3) with small coloured accent dots.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide white background fill
- title at x 31.5, y 15.9, w 34.5, h 5.4 [OpenAI Sans, Medium weight, sentence case; font size 5.43% of slide height; #000000; center aligned]: Approaches to AI coding
- photo at x 5.7, y 28.2, w 26.8, h 34.6 [#A8C8F0]: Card 1 illustration: code complete screenshot in a light blue to lavender gradient rounded card showing a CSS code editor with Copilot suggestion and Accept/Discard buttons (Rounded corners about 2.5 percent of height, soft drop shadow, full card unCropped, inner screenshot about x 5.5 y 25.2 w 30.6 h 40.5 overlapping the card edges)
- photo at x 36.6, y 28.2, w 26.8, h 34.6 [#A8C8F0]: Card 2 illustration: pair programming screenshot in a light blue to lavender gradient rounded card showing a dark IDE with Copilot chat and module.py (Rounded corners about 2.5 percent of height, soft drop shadow, inner dark screenshot about x 37.1 y 25.2 w 29.9 h 40.5 boxed with rounded corners)
- photo at x 67.5, y 28.2, w 26.8, h 34.6 [#4B5BE6]: Card 3 illustration: agentic delegation screenshot in a blue-violet gradient rounded card showing a light app window with a blurred placeholder response and heading Create a snake game (Rounded corners about 2.5 percent of height, soft drop shadow, inner window about x 67.2 y 25.2 w 30.6 h 40.5 with blurred content)
- body at x 8.3, y 67.3, w 13.1, h 11.9 [OpenAI Sans, Regular weight, sentence case; font size 5.43% of slide height; #000000; left aligned]: Code / complete (Two lines, line spacing about 1.0; line 1 box x 8.3 y 67.3 w 7.4 h 5.4, line 2 box x 8.3 y 73.8 w 13.1 h 5.4)
- body at x 39.2, y 67.3, w 19, h 11.9 [OpenAI Sans, Regular weight, sentence case; font size 5.43% of slide height; #000000; left aligned]: Pair / programming (Two lines, line spacing about 1.0; line 1 box x 39.2 y 67.3 w 6.1 h 5.4, line 2 box x 39.2 y 73.8 w 19.0 h 5.4)
- body at x 70.1, y 67.3, w 14.8, h 11.9 [OpenAI Sans, Regular weight, sentence case; font size 5.43% of slide height; #000000; left aligned]: Agentic / delegation (Two lines, line spacing about 1.0; line 1 box x 70.1 y 67.3 w 11.6 h 5.4, line 2 box x 70.1 y 73.8 w 14.8 h 5.4)
- logo at x 2.7, y 94.3, w 5.6, h 1.6 [OpenAI Sans, Bold weight; font size 1.48% of slide height; #000000; left aligned]: OpenAI wordmark (Text logo in bold black, bottom left corner)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans, Regular weight, sentence case; font size 1.48% of slide height; #000000; left aligned]: Conﬁdential and proprietary.

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 5.7, y 28.2, w 26.8, h 54.6: #FFFFFF
- fill at x 67.5, y 28.2, w 26.8, h 54.6: #FFFFFF
- fill at x 36.6, y 28.2, w 26.8, h 54.6: #FFFFFF
- image at x 0.0, y 20.4, w 100.0, h 79.6: 1920 by 1080 pixels (covers the slide)
- image at x 0.0, y 20.3, w 99.8, h 79.7: 1920 by 1080 pixels (covers the slide)
- image at x 0.0, y 21.8, w 100.0, h 78.2: 1920 by 1080 pixels (covers the slide)
- image at x 61.7, y 23.4, w 38.3, h 43.2: 124 by 78 pixels
- image at x 70.8, y 31.3, w 29.2, h 43.2: 495 by 310 pixels
- image at x 62.5, y 19.8, w 35.0, h 38.9: 8 by 5 pixels
- image at x 66.3, y 19.8, w 33.7, h 38.9: 31 by 20 pixels
- image at x 67.2, y 25.2, w 30.6, h 40.5: 441 by 329 pixels
- image at x 5.5, y 25.2, w 30.6, h 40.5: 441 by 329 pixels
- image at x 37.1, y 25.2, w 29.9, h 40.5: 430 by 329 pixels
- image at x 40.6, y 31.3, w 27.2, h 28.5: 512 by 302 pixels
- image at x 9.0, y 31.3, w 23.6, h 28.8: 384 by 256 pixels
- image at x 11.8, y 33.0, w 4.2, h 7.5: 61 by 61 pixels
- text at x 31.5, y 15.9, w 34.5, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: Approaches to AI coding
- text at x 8.3, y 67.3, w 7.4, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Code
- text at x 39.2, y 67.3, w 6.1, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Pair
- text at x 70.1, y 67.3, w 11.6, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Agentic
- text at x 8.3, y 73.8, w 13.1, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: complete
- text at x 39.2, y 73.8, w 19.0, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: programming
- text at x 70.1, y 73.8, w 14.8, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: delegation
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 5

**Text**

One Agent, Everywhere you Work
CLI
Original & Popular UI
Zero-Dependency Install
Powerful Slash Commands
(/compact /fork /new)
Integration with ChatGPT
Apps
Codex App
Easily run tasks in parallel
across multiple workspaces
Better/more flexible diff views
Excellent visual surface for
skills, worktrees &
automations
Great UI for workflows beyond
writing code
IDEs
(VS Code, JetBrains, Xcode)
Fits naturally into IDE-based
workflows
Good for incremental tasks,
quick assistance
Great for ‘chat-style’ usage
Confidential and proprietary.

**Layout**

Title sits top-left at about 4% from the left edge, on a white band across the top about 13% of the slide height, with the blue background starting below it. Three white rounded cards sit in columns: left and right cards are equal height and aligned at the same top and bottom, while the centre card rises about 9% higher than the sides. Each card holds a centred icon and heading followed by left-aligned bullets; the confidentiality line sits at the bottom left.

**Visuals**

Five app icons, all roughly square with soft shadows. Left card: a dark charcoal rounded-square terminal icon with a white '>_' prompt, at about x 13.9 y 28.3 w 6.4 h 11.3 percent, overlapping the top edge of the card. Centre card: a white rounded-square app icon with a blue-violet cloud shape containing a '>_' terminal prompt, at about x 45.9 y 17.5 w 8.7 h 15.4, overlapping the top edge of the card with a drop shadow. Right card: the VS Code logo (blue stylised ribbon) at about x 71.2 y 27.1 w 7.5 h 13.4; a black square JetBrains-style terminal icon with a white underscore at about x 79.6 y 27.3 w 7.3 h 13.0 with a soft grey shadow; and the Xcode icon (light blue rounded square with a hammer and crossed tools) at about x 86.8 y 26.2 w 8.6 h 15.4. The background is a smooth blue gradient cloud photograph with pale periwinkle top-left and saturated blue bottom-right, full-bleed behind everything.

**Design**

Background is a full-bleed blue gradient image with pale periwinkle #C9D6FF at top left shading to deep blue #2B3FE0 at right and bottom, with a white band across the top. Title and headings are black #000000 in OpenAI Sans Regular (title) and OpenAI Sans Bold (headings), a geometric grotesque sans-serif. Title is about 6.9% of slide height; headings about 4.2%; bullet and body text about 3.0%. Bullet glyphs are black solid circles in Arial. Cards are white #FFFFFF with rounded corners of about 1% of slide width and no border; a faint shadow may be present. No accent colours beyond the app icons.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: white base fill covering the whole slide (Lowest layer, covered by the background image.)
- photo at x 0, y 0, w 100, h 100 [#2B3FE0]: blue gradient cloud-sky background illustration, full-bleed (Pale periwinkle top left shading to deep blue right and bottom; white band across top; no overlay or tint; not cropped.)
- shape at x 2.1, y 33.5, w 31.1, h 60.9 [#FFFFFF]: left white card (Rounded corners about 1% of slide width; no stroke; faint shadow; opacity 100%.)
- shape at x 34.4, y 24.8, w 31.1, h 69.6 [#FFFFFF]: centre white card (Rounded corners about 1% of slide width; no stroke; faint shadow; top edge higher than side cards.)
- shape at x 66.8, y 33.5, w 31.1, h 60.9 [#FFFFFF]: right white card (Rounded corners about 1% of slide width; no stroke; faint shadow.)
- icon at x 13.9, y 28.3, w 6.4, h 11.3 [#2A2A2A]: CLI terminal app icon: dark charcoal rounded square with white '>_' prompt (Straddles the top edge of the left card; drop shadow.)
- icon at x 45.9, y 17.5, w 8.7, h 15.4 [#FFFFFF]: Codex app icon: white rounded square with blue-violet cloud containing '>_' prompt (Straddles the top edge of the centre card; soft drop shadow.)
- icon at x 71.2, y 27.1, w 7.5, h 13.4 [#0078D4]: VS Code logo (blue ribbon) (Sits above the right card top edge; no background tile.)
- icon at x 79.6, y 27.3, w 7.3, h 13 [#000000]: JetBrains-style black square terminal icon with white underscore (Soft grey shadow beneath.)
- icon at x 86.8, y 26.2, w 8.6, h 15.4 [#3AA6E8]: Xcode app icon: light blue rounded square with hammer and crossed tools (Sits above the right card top edge; soft shadow.)
- title at x 3.9, y 4.5, w 59.2, h 6.9 [OpenAISans-Regular, regular weight, sentence case, no letter-spacing; font size 6.91% of slide height; #000000; left aligned]: One Agent, Everywhere you Work (Sits on the white band.)
- subtitle at x 15.2, y 40.1, w 3.9, h 4.2 [OpenAISans-Bold, bold, sentence case; font size 4.2% of slide height; #000000; center aligned]: CLI (Heading of left card, centred on the card.)
- subtitle at x 44, y 39.3, w 12.6, h 4.2 [OpenAISans-Bold, bold, sentence case; font size 4.2% of slide height; #000000; center aligned]: Codex App (Heading of centre card, centred on the card.)
- subtitle at x 80.4, y 41.5, w 5.1, h 4.2 [OpenAISans-Bold, bold, sentence case; font size 4.2% of slide height; #000000; center aligned]: IDEs (Heading of right card, centred on the card.)
- body at x 6, y 47.6, w 23.4, h 31.1 [OpenAISans-Regular, regular weight, sentence case; font size 2.96% of slide height; #000000; left aligned]: Original & Popular UI / Zero-Dependency Install / Powerful Slash Commands / (/compact /fork /new) / Integration with ChatGPT / Apps (Bullet list of four items with solid black circle bullets in Arial at x 6.0 and text indented to x 8.9; line spacing about 1.5 within item; about 3.4% between items; items separated by blank space; the bullet list is the left card content.)
- body at x 37.4, y 46.4, w 26.2, h 37.6 [OpenAISans-Regular, regular weight, sentence case; font size 2.96% of slide height; #000000; left aligned]: Easily run tasks in parallel / across multiple workspaces / Better/more flexible diff views / Excellent visual surface for / skills, worktrees & / automations / Great UI for workflows beyond / writing code (Bullet list of four items with solid black circle bullets in Arial at x 37.4 and text indented to x 40.2; line spacing about 1.5 within item; the bullet list is the centre card content.)
- body at x 71.9, y 49.3, w 20.9, h 3 [OpenAISans-Regular, regular weight, sentence case; font size 2.96% of slide height; #000000; left aligned]: (VS Code, JetBrains, Xcode) (Sub-line under the IDEs heading, no bullet.)
- body at x 69.1, y 57.3, w 24.7, h 24.3 [OpenAISans-Regular, regular weight, sentence case; font size 2.96% of slide height; #000000; left aligned]: Fits naturally into IDE-based / workflows / Good for incremental tasks, / quick assistance / Great for ‘chat-style’ usage (Bullet list of three items with solid black circle bullets in Arial at x 69.1 and text indented to x 71.9; line spacing about 1.5 within item; the bullet list is the right card content.)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue, regular weight, sentence case; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Bottom-left confidentiality line on the blue background.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 2.1, y 33.5, w 31.1, h 60.9: #FFFFFF
- fill at x 34.4, y 24.8, w 31.1, h 69.6: #FFFFFF
- fill at x 66.8, y 33.5, w 31.1, h 60.9: #FFFFFF
- image at x 0.0, y 0.0, w 100.0, h 100.0: 1920 by 1080 pixels (covers the slide)
- image at x 45.9, y 17.5, w 8.7, h 15.4: 125 by 125 pixels
- image at x 86.8, y 26.2, w 8.6, h 15.4: 125 by 125 pixels
- image at x 71.2, y 27.1, w 7.5, h 13.4: 109 by 109 pixels
- image at x 46.2, y 18.1, w 7.5, h 13.3: 102 by 102 pixels
- image at x 87.1, y 26.7, w 7.5, h 13.3: 128 by 128 pixels
- image at x 79.6, y 27.3, w 7.3, h 13.0: 106 by 106 pixels
- image at x 14.1, y 28.8, w 6.5, h 11.5: 94 by 94 pixels
- image at x 13.9, y 28.3, w 6.4, h 11.3: 128 by 128 pixels
- image at x 71.5, y 27.7, w 6.4, h 11.3: 128 by 128 pixels
- image at x 80.4, y 30.1, w 5.0, h 8.9: 73 by 73 pixels
- image at x 80.9, y 35.4, w 2.8, h 2.8: 41 by 23 pixels
- text at x 3.9, y 4.5, w 59.2, h 6.9 [OpenAISans-Regular; font size 6.91; #000000]: One Agent, Everywhere you Work
- text at x 15.2, y 40.1, w 3.9, h 4.2 [OpenAISans-Bold; font size 4.20; #000000]: CLI
- text at x 44.0, y 39.3, w 12.6, h 4.2 [OpenAISans-Bold; font size 4.20; #000000]: Codex App
- text at x 80.4, y 41.5, w 5.1, h 4.2 [OpenAISans-Bold; font size 4.20; #000000]: IDEs
- text at x 37.4, y 46.4, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 40.2, y 46.4, w 19.9, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Easily run tasks in parallel
- text at x 6.0, y 47.6, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 8.9, y 47.6, w 15.7, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Original & Popular UI
- text at x 71.9, y 49.3, w 20.9, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: (VS Code, JetBrains, Xcode)
- text at x 40.2, y 50.5, w 21.0, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: across multiple workspaces
- text at x 6.0, y 54.0, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 8.9, y 54.0, w 18.6, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Zero-Dependency Install
- text at x 37.4, y 56.6, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 40.2, y 56.6, w 22.8, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Better/more ﬂexible diff views
- text at x 69.1, y 57.3, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 71.9, y 57.3, w 21.9, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Fits naturally into IDE-based
- text at x 6.0, y 60.4, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 8.9, y 60.4, w 20.5, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Powerful Slash Commands
- text at x 71.9, y 60.9, w 7.8, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: workﬂows
- text at x 37.4, y 62.7, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 40.2, y 62.7, w 20.7, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Excellent visual surface for
- text at x 8.9, y 64.9, w 16.8, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: (/compact /fork /new)
- text at x 40.2, y 66.7, w 13.9, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: skills, worktrees &
- text at x 69.1, y 68.0, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 71.9, y 68.0, w 21.3, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Good for incremental tasks,
- text at x 6.0, y 71.3, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 8.9, y 71.3, w 19.8, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Integration with ChatGPT
- text at x 40.2, y 70.8, w 9.6, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: automations
- text at x 71.9, y 71.5, w 12.5, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: quick assistance
- text at x 8.9, y 75.7, w 4.4, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Apps
- text at x 37.4, y 76.9, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 40.2, y 76.9, w 23.4, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Great UI for workﬂows beyond
- text at x 69.1, y 78.6, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 71.9, y 78.6, w 20.3, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Great for ‘chat-style’ usage
- text at x 40.2, y 81.0, w 9.5, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: writing code
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 6

**Text**

Installation & Auth
CLI
Codex App
IDEs
Initiate login flow: / codex login / To start a session: codex
openai.com/codex/ / Download / Drag into Applications / Sign in with ChatGPT
Open Extensions Pane / Search “OpenAI Codex” / Install / Sign in with ChatGPT
Confidential and proprietary.

**Layout**

Three white rounded cards sit side by side in columns over a full-bleed blue background, with the title left-aligned in the white band at the top. The left card (CLI) and right card (IDEs) are the same height and sit slightly lower, while the centre card (Codex App) is taller and rises higher. Each card has a centred bold heading below its icon, and bullet or text content is left-aligned inside. Margins are about 2% at the sides and the cards end near 94% of the slide height. No slide number is visible; a small confidentiality line appears at the bottom left.

**Visuals**

Three app icons sit on top of the cards: a dark pixelated CLI terminal icon above the CLI card, a pixelated blue-violet Codex app icon above the centre card, and, above the right card, a blue pixelated VS Code icon, a black terminal icon with a soft shadow, and a blue Xcode icon with a hammer. Inside the CLI card is a dark navy code panel with five numbered lines: 1 npm install -g @openai/codex, 2 blank, 3 # or, 4 blank, 5 brew install --cask codex, with the keywords npm and brew in green. Below the IDE bullets is a dark screenshot of the Codex extension in VS Code showing a Codex logo, the title Codex – OpenAI's coding, an Install/Disable/Uninstall button row, and an Installation panel. The screenshot is boxed with a dark shadow and is not cropped. The background is a blue-violet gradient photograph with soft white cloud-like haze at upper left.

**Design**

Background: Full-bleed blue-violet gradient image (#7F9BFF to #2B3FE0) with a white band across the top from 0 to 13.5% of slide height. The top band is white #FFFFFF and holds a large black sans-serif title. The lower background is a blue-violet gradient, roughly #7F9BFF at the upper left shading to #2B3FE0 at the right and bottom. Titles and bullets are black #000000. Card headings are bold black. The link and accent colours are teal #0097A7 for openai.com/codex/ and green #6AA84F for code text. Typeface is OpenAI Sans in Regular and Bold, a geometric sans-serif. Title height is about 6.9% of slide height and card headings about 4.2%. Body text is about 3.0% and the code text is about 3.0% in Courier New Bold. Cards are white with rounded corners of about 1.5% of slide width, a soft drop shadow, and no border.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Background: full-bleed blue-violet gradient image covering the slide, with white band at top (Gradient runs from #7F9BFF upper left to #2B3FE0 lower right with soft white haze at upper left; top band white from y 0 to 13.5%.)
- title at x 3.9, y 4.5, w 32.7, h 6.9 [OpenAI Sans Regular; font size 6.91% of slide height; #000000; left aligned]: Installation & Auth (Sits in the white top band; single line.)
- shape at x 2.1, y 33.5, w 31.1, h 60.9 [#FFFFFF]: Left white card (CLI) (Rounded corners about 1.5% of width, soft drop shadow, no border.)
- shape at x 34.4, y 24.8, w 31.1, h 69.6 [#FFFFFF]: Centre white card (Codex App) (Rounded corners about 1.5% of width, soft drop shadow, no border; rises above the side cards.)
- shape at x 66.8, y 33.5, w 31.1, h 60.9 [#FFFFFF]: Right white card (IDEs) (Rounded corners about 1.5% of width, soft drop shadow, no border.)
- icon at x 14.1, y 28.8, w 6.5, h 11.5 [#2A2A2A]: CLI terminal app icon: dark grey pixelated rounded square with a small white cursor glyph (Sits centred on the top edge of the CLI card; rounded square about 20% corner radius; soft shadow.)
- icon at x 45.9, y 17.5, w 8.7, h 15.4 [#8E8CF0]: Codex app icon: pixelated rounded square with blue-violet gradient and a glossy blue lower area (Sits centred on the top edge of the centre card; rounded square; soft shadow.)
- icon at x 71.2, y 27.1, w 7.5, h 13.4 [#1E8BD6]: VS Code icon: blue pixelated interlaced ribbon mark (Sits above the top edge of the right card; no background tile.)
- icon at x 80.4, y 30.1, w 5, h 8.9 [#000000]: Terminal app icon: black square with a white underline cursor (Square with soft grey drop shadow beneath and to the right.)
- icon at x 86.8, y 26.2, w 8.6, h 15.4 [#2E9FE0]: Xcode app icon: blue rounded square with a hammer and a stripe (Rounded square with soft shadow; hammer diagonal in dark grey.)
- heading at x 15.2, y 40.1, w 3.9, h 4.2 [OpenAI Sans Bold; font size 4.2% of slide height; #000000; center aligned]: CLI
- illustration at x 4.9, y 45.2, w 25.5, h 17 [Courier New Regular; font size 2.96% of slide height; #232B3B; left aligned]: Dark code panel with line numbers 1 to 5: npm install -g @openai/codex / (blank) / # or / (blank) / brew install --cask codex (Rounded rectangle, corner radius about 1%; line numbers in grey #6B7280; npm and brew in green #3FB950; other text white #F0F0F0; no shadow.)
- body at x 6.1, y 67.6, w 18, h 12 [OpenAI Sans Regular with Courier New Bold code runs; font size 2.96% of slide height; #000000; left aligned]: Initiate login flow: / codex login / To start a session: codex (Line 1 Initiate login flow: in OpenAI Sans Regular black; line 2 codex login in Courier New Bold #6AA84F; line 3 To start a session: in OpenAI Sans Regular black followed by codex in Courier New Bold #6AA84F.)
- heading at x 44, y 39.3, w 12.6, h 4.2 [OpenAI Sans Bold; font size 4.2% of slide height; #000000; center aligned]: Codex App
- bullet list at x 38.3, y 51.3, w 19.8, h 24.3 [OpenAI Sans Regular; font size 2.96% of slide height; #000000; left aligned]: openai.com/codex/ / Download / Drag into Applications / Sign in with ChatGPT (Bullets are solid black round dots in Arial, about 1 unit wide; line pitch about 7.1% of slide height; the first item is an underlined link in #0097A7.)
- heading at x 80.4, y 41.5, w 5.1, h 4.2 [OpenAI Sans Bold; font size 4.2% of slide height; #000000; center aligned]: IDEs
- bullet list at x 70.5, y 51.4, w 23, h 20.7 [OpenAI Sans Regular; font size 3.21% of slide height; #000000; left aligned]: Open Extensions Pane / Search “OpenAI Codex” / Install / Sign in with ChatGPT (Solid black round bullets in Arial; line pitch about 5.8% of slide height.)
- photo at x 73.1, y 73, w 26.9, h 19.5 [#1E1E1E]: Screenshot of the Codex extension page in VS Code: Codex logo, title Codex – OpenAI's coding, OpenAI publisher line, Disable, Uninstall and Switch to Release buttons, and an Installation panel with identifier openai.chatgpt (Boxed, not cropped; dark UI with activity bar on left; blue primary button; soft shadow; placed over the lower right of the IDEs card and slightly overlaps its bottom edge.)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [Helvetica Neue Regular; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Listed in the file measurements; not clearly legible in the rendered image.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 2.1, y 33.5, w 31.1, h 60.9: #FFFFFF
- fill at x 66.8, y 33.5, w 31.1, h 60.9: #FFFFFF
- fill at x 34.4, y 24.8, w 31.1, h 69.6: #FFFFFF
- image at x 0.0, y 0.0, w 100.0, h 100.0: 1920 by 1080 pixels (covers the slide)
- image at x 0.8, y 38.0, w 33.7, h 31.5: 512 by 270 pixels
- image at x 73.1, y 73.0, w 26.9, h 19.5: 512 by 180 pixels
- image at x 72.6, y 73.6, w 19.5, h 20.1: 281 by 163 pixels
- image at x 45.9, y 17.5, w 8.7, h 15.4: 125 by 125 pixels
- image at x 86.8, y 26.2, w 8.6, h 15.4: 125 by 125 pixels
- image at x 71.2, y 27.1, w 7.5, h 13.4: 109 by 109 pixels
- image at x 46.2, y 18.1, w 7.5, h 13.3: 13 by 13 pixels
- image at x 87.1, y 26.7, w 7.5, h 13.3: 32 by 32 pixels
- image at x 79.6, y 27.3, w 7.3, h 13.0: 106 by 106 pixels
- image at x 14.1, y 28.8, w 6.5, h 11.5: 94 by 94 pixels
- image at x 13.9, y 28.3, w 6.4, h 11.3: 16 by 16 pixels
- image at x 71.5, y 27.7, w 6.4, h 11.3: 8 by 8 pixels
- image at x 80.4, y 30.1, w 5.0, h 8.9: 73 by 73 pixels
- image at x 80.9, y 35.4, w 2.8, h 2.8: 41 by 23 pixels
- text at x 3.9, y 4.5, w 32.7, h 6.9 [OpenAISans-Regular; font size 6.91; #000000]: Installation & Auth
- text at x 15.2, y 40.1, w 3.9, h 4.2 [OpenAISans-Bold; font size 4.20; #000000]: CLI
- text at x 44.0, y 39.3, w 12.6, h 4.2 [OpenAISans-Bold; font size 4.20; #000000]: Codex App
- text at x 80.4, y 41.5, w 5.1, h 4.2 [OpenAISans-Bold; font size 4.20; #000000]: IDEs
- text at x 70.5, y 51.4, w 1.1, h 3.2 [ArialMT; font size 3.21; #000000]: ●
- text at x 73.4, y 51.4, w 18.5, h 3.2 [OpenAISans-Regular; font size 3.21; #000000]: Open Extensions Pane
- text at x 38.3, y 51.3, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 41.2, y 51.3, w 14.7, h 3.0 [OpenAISans-Regular; font size 2.96; #0097A7]: openai.com/codex/
- text at x 70.5, y 57.3, w 1.1, h 3.2 [ArialMT; font size 3.21; #000000]: ●
- text at x 73.4, y 57.3, w 20.1, h 3.2 [OpenAISans-Regular; font size 3.21; #000000]: Search “OpenAI Codex”
- text at x 38.3, y 58.4, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 41.2, y 58.4, w 7.6, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Download
- text at x 70.5, y 63.1, w 1.1, h 3.2 [ArialMT; font size 3.21; #000000]: ●
- text at x 73.4, y 63.1, w 5.4, h 3.2 [OpenAISans-Regular; font size 3.21; #000000]: Install
- text at x 70.5, y 68.9, w 1.1, h 3.2 [ArialMT; font size 3.21; #000000]: ●
- text at x 73.4, y 68.9, w 17.6, h 3.2 [OpenAISans-Regular; font size 3.21; #000000]: Sign in with ChatGPT
- text at x 38.3, y 65.5, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 41.2, y 65.5, w 16.9, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Drag into Applications
- text at x 6.1, y 67.6, w 13.9, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Initiate login ﬂow:
- text at x 6.1, y 71.5, w 10.0, h 3.0 [CourierNew-Bold; font size 2.96; #6AA84F]: c o d e x l o g i n
- text at x 38.3, y 72.6, w 1.0, h 3.0 [ArialMT; font size 2.96; #000000]: ●
- text at x 41.2, y 72.6, w 16.3, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: Sign in with ChatGPT
- text at x 6.1, y 76.6, w 13.6, h 3.0 [OpenAISans-Regular; font size 2.96; #000000]: To start a session:
- text at x 20.7, y 76.6, w 4.0, h 3.0 [CourierNew-Bold; font size 2.96; #6AA84F]: c o d e x
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 7

**Text**

Use Codex on one of your / repos or an open source / repo / Need a repo? A few options / Link: github.com/agentsmd/agents.md / Link: / github.com/greenleaf-oai/3js-follow-along-demo / (Advanced) Link: github.com/openai/codex / OpenAI / Confidential and proprietary. / 7

**Layout**

Two-column split slide. The left white column holds a left-aligned title at the top, a bold heading and three link lines in the lower-middle area, all starting at about x 4% with a left margin of about 4%. The right column (x 51.3% to 100%) is a full-height blue gradient panel with a rounded-corner card illustration centred in it. The OpenAI wordmark sits bottom-left with the confidentiality line beside it, and the page number sits at the bottom-right of the slide near y 94%.

**Visuals**

Full-height right-hand panel (x 51.3% to 100%, full slide height) with a soft blue-violet gradient, lighter at the top-right corner. Centred on this panel is a tall rounded-rectangle card (x 56.0%, y 6.2%, w 38.9%, h 80.0%) in deep blue with a thin grey outline. The card's top third is a pale lavender-white gradient sky; below it a deep blue (#2440B8) body begins with a flat-topped, slightly peaked horizon shape. Three small dark indigo squares (#2D1F9E) are stacked diagonally near the upper middle, one larger dark indigo cube (#2D1F9E) sits in the middle-right, and a white rounded-square app icon (about 14% of card width) sits in the lower-middle with a violet-blue cloud-like symbol containing a white dash and chevron. The composition is a centred, unbroken card with no cropping; the image is boxed by the card's rounded corners over the gradient.

**Design**

Background: Split background: solid white #FFFFFF on the left half and a blue-violet gradient photograph (#5B5CF0 to #4F7BF0 with #F0EEFF glow) on the right half. Split layout with a plain white (#FFFFFF) left half and a blue-violet gradient photo panel on the right. Title and body text are black (#000000). Title typeface is a geometric grotesque sans-serif (OpenAI Sans Regular), roughly 5.9% of slide height per line with generous line spacing (about 7% of slide height between baselines). Body is OpenAI Sans Regular at about 3.5% of slide height, with the heading in OpenAI Sans Bold. Links are black, underlined, with no accent colour. Footer uses a small Helvetica Neue regular at about 1.5% of slide height. The illustration card has a thin grey (#8A8FA0) outline about 1 px wide, rounded corners of roughly 12% of card width, and no shadow. Gradient panel runs from violet #6A6BF5 at left to blue #4F7BF0 at bottom-right, with a pale lavender-white #F0EEFF glow at top-right.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: full-slide white background rectangle (Solid fill covering the entire slide, no stroke, behind all other elements.)
- photo at x 51.3, y 0, w 48.7, h 100 [#5B5CF0]: blue-violet gradient background panel, right half of slide, no visible subject (Full-bleed to top, bottom and right edges, no border; soft gradient from violet #6A6BF5 to blue #4F7BF0 with pale lavender-white #F0EEFF glow at top-right.)
- illustration at x 56, y 6.2, w 38.9, h 80 [#2440B8]: rounded-rectangle card with pale sky, deep blue body, indigo squares, cube and app icon with cloud symbol (Rounded corners of about 12% of card width, thin grey outline about 1 px (#8A8FA0), no shadow, opaque; the app icon is a white rounded square with a violet-blue cloud symbol, a white dash and chevron inside; overlaps the gradient panel.)
- title at x 3.9, y 5.1, w 39.4, h 20.1 [OpenAI Sans, Regular, sentence case; font size 5.93% of slide height; #000000; left aligned]: Use Codex on one of your / repos or an open source / repo (Three lines, line spacing about 7% of slide height between baselines.)
- body at x 4.4, y 37.7, w 24.7, h 3.5 [OpenAI Sans, Bold, sentence case; font size 3.46% of slide height; #000000; left aligned]: Need a repo? A few options (Single line heading.)
- body at x 4.4, y 44.7, w 34.8, h 3.5 [OpenAI Sans, Regular, sentence case; font size 3.46% of slide height; #000000; left aligned]: Link: github.com/agentsmd/agents.md (Domain text underlined, black; 'Link:' not underlined.)
- body at x 4.4, y 51.8, w 4.8, h 3.5 [OpenAI Sans, Regular, sentence case; font size 3.46% of slide height; #000000; left aligned]: Link: (Label line above the URL line.)
- body at x 4.4, y 56.3, w 43.2, h 3.5 [OpenAI Sans, Regular, sentence case; font size 3.46% of slide height; #000000; left aligned]: github.com/greenleaf-oai/3js-follow-along-demo (Underlined hyperlink text, black.)
- body at x 4.4, y 63.4, w 38.5, h 3.5 [OpenAI Sans, Regular, sentence case; font size 3.46% of slide height; #000000; left aligned]: (Advanced) Link: github.com/openai/codex (Only the URL github.com/openai/codex is underlined.)
- logo at x 2.7, y 94.2, w 3, h 1.2 [#000000]: OpenAI wordmark (Small bold black wordmark, bottom-left corner.)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [Helvetica Neue, Regular, sentence case; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Single line beside the wordmark.)
- page number at x 96.8, y 94.4, w 0.5, h 1.5 [Helvetica Neue, Regular; font size 1.48% of slide height; #000000; right aligned]: 7 (Bottom-right, slide number 7 of 37.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 51.3, y 0.0, w 48.7, h 100.0: 1920 by 1080 pixels (covers the slide)
- image at x 56.0, y 6.2, w 38.9, h 80.0: 443 by 512 pixels
- text at x 3.9, y 5.1, w 39.4, h 5.9 [OpenAISans-Regular; font size 5.93; #000000]: Use Codex on one of your
- text at x 3.9, y 12.2, w 37.2, h 5.9 [OpenAISans-Regular; font size 5.93; #000000]: repos or an open source
- text at x 3.9, y 19.3, w 6.9, h 5.9 [OpenAISans-Regular; font size 5.93; #000000]: repo
- text at x 4.4, y 37.7, w 24.7, h 3.5 [OpenAISans-Bold; font size 3.46; #000000]: Need a repo? A few options
- text at x 4.4, y 44.7, w 34.8, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Link: github.com/agentsmd/agents.md
- text at x 4.4, y 51.8, w 4.8, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Link:
- text at x 4.4, y 56.3, w 43.2, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: github.com/greenleaf-oai/3js-follow-along-demo
- text at x 4.4, y 63.4, w 38.5, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: (Advanced) Link: github.com/openai/codex
- text at x 96.8, y 94.4, w 0.5, h 1.5 [HelveticaNeue; font size 1.48; #000000]: 7
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 8

**Review notes**

- Review: Some slides also add little, such as the bare gradient on 9 and the sparse slide 8.

**Text**

Send your first prompt to Codex
take a look at the last month of commits and come up with a list of top features we can build next
GPT-5.3-Codex
Extra High
Local
Default permissions
master
OpenAI
Confidential and proprietary.
8

**Layout**

Single-column composition on a plain white slide. The title sits top-left at the left margin, about 7.6% from the top. A dark rounded app-style panel is centred horizontally in the middle band of the slide, spanning roughly 6% to 94% of the width and 34% to 66% of the height. The footer runs along the bottom: the OpenAI wordmark at the left margin, the confidentiality line next to it, and a tiny page number at the right margin, all on one baseline near 94.5% of the height.

**Visuals**

One photographic screenshot of a dark Codex app prompt box, placed at 6.1%, 34.0%, 87.9% by 32.1% of the slide, with no crop or border beyond its own rounded corners. Subject: a chat-style input composer. The outer panel is near-black (#161616) with large rounded corners (about 6% of its height). Inside it sits a slightly lighter (#1E1E1E) rounded input field with a thin grey (#333333) outline, occupying about 7.9% to 92.3% of the slide width and 38.2% to 57.8% of its height. Inside the field, the typed prompt 'take a look at the last month of commits and come up with a list of top features we can build next' is set in light grey-white (#F0F0F0) sans-serif about 2.2% of slide height, followed by a thin white text cursor at its end. Below the prompt, along the bottom of the field, are grey (#8A8A8A) controls: a plus icon at the left, the model selector 'GPT-5.3-Codex' with a downward chevron, and the effort selector 'Extra High' with a downward chevron; at the right are a grey microphone icon and a white circular send button (#FFFFFF, about 4.5% of slide width) containing a black up-arrow. Below the field, outside it but inside the dark panel, are grey controls: a laptop icon with 'Local' and a chevron, a shield-style icon with 'Default permissions' and a chevron, at the right a git-branch icon with 'master' and a chevron, and a small grey loading spinner ring at the far right. Lighting is flat, even UI lighting with no shadows. Dominant colours are near-black and grey with a white send button. The image is a screenshot, not a photograph. It is not cropped and is not full-bleed. A yellow-green (#EEFF41) fill is recorded in the file at the same box as the screenshot; in the rendered slide that area reads as the dark panel with white space around its rounded corners.

**Design**

Background: Solid white #FFFFFF. Clean white corporate slide. Background white #FFFFFF. Title in black #000000, regular-weight geometric sans-serif (OpenAI Sans Regular), about 5.9% of slide height, sentence case, left aligned. Footer text in black #000000, Helvetica Neue regular, about 1.5% of slide height. Page number in black #000000, Arial regular, about 0.6% of slide height. Accent colour: the dark panel #161616 with a white send button; the file also records a yellow-green #EEFF41 fill at the screenshot box. No gradients, no visible shadows on the slide itself. The panel has large rounded corners and the input field has smaller rounded corners.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 6.1, y 34, w 87.9, h 32.1 [#EEFF41]: Yellow-green filled rectangle recorded in the file at the screenshot box, sitting behind the screenshot (Fill only; no stroke; covered by the screenshot in the render.)
- photo at x 6.1, y 34, w 87.9, h 32.1 [#161616]: Screenshot of the Codex dark prompt composer showing the typed prompt 'take a look at the last month of commits and come up with a list of top features we can build next', model selector 'GPT-5.3-Codex', effort selector 'Extra High', 'Local', 'Default permissions', 'master' and icons (plus, microphone, send arrow, laptop, shield, git branch, spinner) (Image 1696 by 348 pixels, unboxed, rounded outer panel with corner radius about 6% of its height, transparent corners showing white slide background; inner input field rounded with thin #333333 outline; white circular send button about 4.5% of slide width.)
- title at x 3.9, y 7.6, w 48.7, h 5.9 [OpenAI Sans, Regular, sentence case, no italics, no caps, normal letter-spacing; font size 5.93% of slide height; #000000; left aligned]: Send your first prompt to Codex
- logo at x 2.7, y 94.3, w 3, h 1.4 [#000000]: OpenAI wordmark (Bold black sans-serif wordmark at bottom left.)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [Helvetica Neue, Regular, sentence case; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.
- page number at x 96.2, y 94.8, w 0.2, h 0.6 [Arial, Regular; font size 0.62% of slide height; #000000; right aligned]: 8

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 6.1, y 34.0, w 87.9, h 32.1: #EEFF41
- image at x 6.1, y 34.0, w 87.9, h 32.1: 1696 by 348 pixels
- text at x 3.9, y 7.6, w 48.7, h 5.9 [OpenAISans-Regular; font size 5.93; #000000]: Send your ﬁrst prompt to Codex
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.
- text at x 96.2, y 94.8, w 0.2, h 0.6 [ArialMT; font size 0.62; #000000]: 8

### Slide 9

**Text**

Q&A
OpenAI

**Layout**

Single-column composition on a white field: the title "Q&A" sits top-left at the margin and the OpenAI wordmark sits top-right on the same header band. A wide full-width image block fills the lower part of the slide between left and right margins of about 2.7% and sits below the header with a gap; no footer or slide number is visible.

**Visuals**

One photograph-like abstract image: a soft, out-of-focus blue and violet gradient surface with a diagonal light-to-dark transition running from the upper-left (pale white-blue, fine streaked texture) to the lower-right (saturated indigo-blue #2E35D6 and violet). The upper-right edge shows a light lavender highlight. The image is boxed (not full-bleed) with small rounded corners, full colour, no overlay, no border, no text. The OpenAI wordmark in black is a logo at top-right.

**Design**

Background: Solid white #FFFFFF. White background #FFFFFF with black #000000 text. Title and wordmark are black #000000. Title typeface is a geometric grotesque sans-serif (Arial-like) in regular weight, approximately 8.9% of slide height for capitals. Wordmark is a bold geometric sans-serif black logotype. Image has rounded corners of about 0.5% of slide height, no shadow, no gradient on the frame itself.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Background fill, white field covering the slide (Solid fill, no stroke, covers the entire slide behind all other elements.)
- photo at x 2.7, y 17, w 94.6, h 83 [#4A5BE8]: Abstract blue-violet gradient photograph: pale white-blue upper-left fading to saturated indigo-blue and violet lower-right, soft focus, boxed with rounded corners (Placed image 1920 by 1080 px, rounded corner radius about 0.5% of slide height, no border, no shadow, no overlay, not cropped to a shape other than the rectangle, dominant colours white #FFFFFF, periwinkle #7F95FF, royal blue #3D5AF1, indigo #2E35D6.)
- title at x 3.6, y 5.8, w 10.5, h 8.9 [ArialMT, regular; font size 8.89% of slide height; #000000; left aligned]: Q&A (Single line, no italics, no capitals styling beyond the letters themselves, normal letter-spacing.)
- logo at x 88.8, y 4.5, w 8, h 4 [#000000; right aligned]: OpenAI wordmark (Black bold geometric sans-serif wordmark at top-right, aligned to the same header band as the title, right margin about 2.2%.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 2.7, y 17.0, w 94.6, h 83.0: 1920 by 1080 pixels
- text at x 3.6, y 5.8, w 10.5, h 8.9 [ArialMT; font size 8.89; #000000]: Q&A

### Slide 10

**Text**

OpenAI
Understanding
Agents
What is the Agent Loop?

**Layout**

Two-column split on a white 16:9 slide. The text column sits on the left with the OpenAI wordmark at top-left (about 2.7% from the left and 4.6% from the top), then the title in two lines and the blue subtitle below it, all left-aligned with no footer or slide number. The square-cornered image panel with rounded corners sits on the right, starting at the top margin of about 4.8% and running to about 95.5% of the slide height.

**Visuals**

Photograph or illustration: a soft gradient background panel with a rounded-corner rectangle, shifting from pale lavender-white at top-left through periwinkle to saturated blue and indigo at bottom-right, with a faint purple glow in the lower centre. Centred on it is a single app-style icon: a light grey-white rounded-square tile (about #F2F2F5) holding a glossy blue-to-violet scalloped cloud shape (violet #B07CF0 at top fading to blue #2F3FE0 at bottom) with a white chevron '>' and a white horizontal dash inside, like a terminal prompt. The icon is not cropped, has a soft drop shadow, and sits in the middle of the panel. The panel has no overlay or tint beyond the photo itself.

**Design**

Background: Solid white #FFFFFF fill covering the whole slide. Plain white background (#FFFFFF). Title text is black (#000000) in an Arial-style regular sans-serif at about 10.1% of slide height (cap height about 7% of slide height), with tight default leading. The subtitle is in OpenAI Sans SemiBold, a geometric sans-serif, at about 4.2% of slide height, in cobalt blue #3442E0. The wordmark is black. The only accent colour is the blue #3442E0 of the subtitle, which echoes the blue image. Image panel has small rounded corners (about 1% of width) and no border.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide background fill, plain white (Covers the full slide, no border, no shadow.)
- photo at x 24.4, y 4.8, w 75.6, h 90.7 [#3A5CF0]: Blue-violet gradient panel with soft light and rounded corners, no text (Placed image 1920 by 1080 px, shown as a rounded-corner rectangle with corner radius about 1% of slide width. Measured box is as read from the file; visually the panel occupies the right side of the slide from about x 56.8 to 97.3 with the same vertical extent. Not cropped beyond the rounded corners, no overlay, no shadow.)
- icon at x 70, y 37.4, w 14.1, h 25.1 [#F2F2F5]: Terminal-style app icon: white rounded-square tile with a blue-violet cloud shape containing a white chevron and dash (Placed image 269 by 269 px, centred on the photo panel. Light grey-white tile with rounded corners, glossy blue-violet scalloped cloud inside, white chevron '>' and white dash, soft drop shadow beneath the tile, no rotation, no opacity change, sits over the photo.)
- logo at x 2.7, y 4.6, w 7.9, h 3.9 [#000000]: OpenAI wordmark (Black wordmark in the top-left corner, aligned to the left margin of the title.)
- title at x 2.9, y 14.1, w 38.6, h 10.1 [ArialMT, regular, sentence case, no italics, no letter-spacing; font size 10.12% of slide height; #000000; left aligned]: Understanding (First line of the title, line spacing about 1.1 of size.)
- title at x 2.9, y 26.2, w 17.7, h 10.1 [ArialMT, regular, sentence case, no italics, no letter-spacing; font size 10.12% of slide height; #000000; left aligned]: Agents (Second line of the title, sits directly under 'Understanding' with the same leading.)
- subtitle at x 2.9, y 40.8, w 26.9, h 4.2 [OpenAISans-SemiBold, semibold, sentence case, no italics, no letter-spacing; font size 4.2% of slide height; #3442E0; left aligned]: What is the Agent Loop? (Single line, sits about 8% of slide height below the title.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 24.4, y 4.8, w 75.6, h 90.7: 1920 by 1080 pixels
- image at x 70.0, y 37.4, w 14.1, h 25.1: 269 by 269 pixels
- text at x 2.9, y 14.1, w 38.6, h 10.1 [ArialMT; font size 10.12; #000000]: Understanding
- text at x 2.9, y 26.2, w 17.7, h 10.1 [ArialMT; font size 10.12; #000000]: Agents
- text at x 2.9, y 40.8, w 26.9, h 4.2 [OpenAISans-SemiBold; font size 4.20; #3442E0]: What is the Agent Loop?

### Slide 11

**Review notes**

- Worth studying: A simple flow diagram of the agent loop shows the process from prompt to tool calls to completion in a single visual.

**Text**

Codex Agentic Loop
It starts with a prompt
Give the agent initial direction which files
to read and what your goals are
Agent calls Tools
Reads your code, Searches company
knowledge, Writes to a file, loops until task
completion
Builds the conversation history
Every action taken & every file read builds
the models context
USER INPUT
MODEL INFERENCE
AGENT RESPONSE
TOOL CALLS
OpenAI
Confidential and proprietary.
11

**Layout**

Two-column slide on a white field: a left text column starting at about 4.9% from the left edge holds the title above three bold headings each followed by a body paragraph, and a right-hand diagram sits in the right half, vertically centred around the middle of the slide. The title sits top-left at about 6% from the top, left-aligned with the text column. A footer row near the bottom has the OpenAI wordmark at left, the confidentiality line beside it, and the small slide number at bottom right.

**Visuals**

One flow diagram, boxed by no frame, on white. Shapes: a top pill-shaped (stadium) box labelled USER INPUT in monospace capitals, filled with a very pale green dotted/stippled tint and a dark olive-green outline; below it a down arrow connects to a rectangle labelled MODEL INFERENCE, pale green fill (#D9EFC4 approx.) with thin blue-grey vertical bars at its left and right ends and a blue-grey outline; from the bottom of MODEL INFERENCE a connector splits into two elbow arrows pointing down: the left one ends in a green pill-shaped box labelled AGENT RESPONSE filled solid green (#A3D977 approx.) with a dark green outline, the right one ends in a pale green rectangle labelled TOOL CALLS with blue-grey end bars. From the right side of TOOL CALLS a line runs right, turns up and returns left into MODEL INFERENCE with an arrowhead pointing left at its right edge, forming a feedback loop. A thin light grey horizontal baseline runs under AGENT RESPONSE and TOOL CALLS. No chart, table, photo or logo.

**Design**

Background: Solid white #FFFFFF. White background. Title in a geometric grotesque sans-serif (OpenAI Sans style), regular weight, black #000000, large (about 5.9% of slide height for capitals approx. 4.1% cap height). Body headings bold sans-serif black #000000 at about 3.95% of slide height; body paragraphs regular sans-serif black #000000 at about 3.7% of slide height with tight line spacing around 4.2%. Italic used for the words 'loops' and 'context'. Accent colours are green (#A3D977 and pale #D9EFC4) and blue-grey outlines in the diagram; the diagram uses monospace capitals for labels. Shapes have fully rounded ends on pills and small square corners on rectangles, no shadows, no gradients. Footer text is tiny: wordmark bold sans-serif, confidentiality line in HelveticaNeue at about 1.5% of height, slide number in Arial at about 0.6%.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- title at x 4.4, y 6.1, w 30.8, h 5.9 [OpenAISans-Regular, regular weight, sentence case; font size 5.93% of slide height; #000000; left aligned]: Codex Agentic Loop (Single line, sits above the diagram and left column.)
- body at x 4.9, y 23.4, w 23.2, h 4 [OpenAISans-Bold, bold, sentence case; font size 3.95% of slide height; #000000; left aligned]: It starts with a prompt (Bold heading of first block.)
- body at x 4.9, y 30.3, w 39.5, h 7.9 [OpenAISans-Regular, regular weight; font size 3.7% of slide height; #000000; left aligned]: Give the agent initial direction which files / to read and what your goals are (Two lines, line spacing about 4.2% of slide height. Only text box with line break shown as slash.)
- body at x 4.9, y 41.3, w 17.5, h 4 [OpenAISans-Bold, bold, sentence case; font size 3.95% of slide height; #000000; left aligned]: Agent calls Tools (Bold heading of second block.)
- body at x 4.9, y 48.2, w 35.3, h 12 [OpenAISans-Regular, regular weight with italic 'loops'; font size 3.7% of slide height; #000000; left aligned]: Reads your code, Searches company / knowledge, Writes to a file, loops until task / completion (Three lines. The word 'loops' is italic (OpenAISans-Italic) at x 30.7, y 52.4.)
- body at x 4.9, y 63.4, w 32.3, h 4 [OpenAISans-Bold, bold, sentence case; font size 3.95% of slide height; #000000; left aligned]: Builds the conversation history (Bold heading of third block.)
- body at x 4.9, y 70.3, w 40.1, h 7.9 [OpenAISans-Regular, regular weight with italic 'context'; font size 3.7% of slide height; #000000; left aligned]: Every action taken & every file read builds / the models context (Two lines. The word 'context' is italic at x 16.0, y 74.5.)
- diagram at x 43, y 16.4, w 55.8, h 45.8 [monospace capitals for labels; #D9EFC4; center aligned]: Flow diagram: USER INPUT pill, MODEL INFERENCE bar, AGENT RESPONSE pill, TOOL CALLS bar, with arrows and feedback loop (Image placed as a raster, 596 by 275 pixels. USER INPUT: stadium shape, x about 59.2 to 81.1 percent of width of slide, y about 27 to 37 percent, very pale green stippled fill, olive-green outline. Downward arrow from its bottom centre to MODEL INFERENCE. MODEL INFERENCE: rectangle, pale green fill #D9EFC4, blue-grey outline with vertical bars near both ends, approx. x 59.2 to 81.1, y 40.5 to 46.7 percent. Split connector from its bottom centre to two downward elbow arrows. AGENT RESPONSE: solid green pill #A3D977 with dark green outline, approx. x 49.6 to 68.4, y 54.5 to 60.7 percent. TOOL CALLS: pale green rectangle with blue-grey end bars, approx. x 73.5 to 92.2, y 54.5 to 60.7 percent. Feedback line from right edge of TOOL CALLS goes right, up and left into right edge of MODEL INFERENCE with a leftward arrowhead. Thin light grey baseline under the bottom row.)
- footer at x 2.6, y 93.6, w 5.5, h 1.5 [OpenAISans-Bold, bold, sentence case; font size 1.5% of slide height; #000000; left aligned]: OpenAI (Wordmark logo rendered as text at bottom left.)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue, regular weight; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Confidentiality line next to wordmark.)
- page number at x 96, y 94.8, w 0.4, h 0.6 [ArialMT, regular weight; font size 0.62% of slide height; #000000; right aligned]: 11 (Bottom right slide number.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 43.0, y 16.4, w 55.8, h 45.8: 596 by 275 pixels
- text at x 4.4, y 6.1, w 30.8, h 5.9 [OpenAISans-Regular; font size 5.93; #000000]: Codex Agentic Loop
- text at x 4.9, y 23.4, w 23.2, h 4.0 [OpenAISans-Bold; font size 3.95; #000000]: It starts with a prompt
- text at x 4.9, y 30.3, w 39.5, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Give the agent initial direction which ﬁles
- text at x 4.9, y 34.5, w 29.8, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: to read and what your goals are
- text at x 4.9, y 41.3, w 17.5, h 4.0 [OpenAISans-Bold; font size 3.95; #000000]: Agent calls Tools
- text at x 4.9, y 48.2, w 35.3, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Reads your code, Searches company
- text at x 4.9, y 52.4, w 25.8, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: knowledge, Writes to a ﬁle,
- text at x 30.7, y 52.4, w 5.2, h 3.7 [OpenAISans-Italic; font size 3.70; #000000]: loops
- text at x 35.8, y 52.4, w 9.7, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: until task
- text at x 4.9, y 56.7, w 10.7, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: completion
- text at x 4.9, y 63.4, w 32.3, h 4.0 [OpenAISans-Bold; font size 3.95; #000000]: Builds the conversation history
- text at x 4.9, y 70.3, w 40.1, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Every action taken & every ﬁle read builds
- text at x 4.9, y 74.5, w 11.1, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: the models
- text at x 16.0, y 74.5, w 7.2, h 3.7 [OpenAISans-Italic; font size 3.70; #000000]: context
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.
- text at x 96.0, y 94.8, w 0.4, h 0.6 [ArialMT; font size 0.62; #000000]: 11

### Slide 12

**Text**

What Is Context?
Why does it matter?
The Agent’s Working Memory
Context is the total information an agent has to complete its task
The Context Window
Codex can hold hundreds of documents in context at once (400-600 pages of text)
Why it Matters
Focused context leads to safer, more accurate, and more consistent agent results
OpenAI
Confidential and proprietary.
12

**Layout**

Two-column split slide. The left white column (about 0 to 51% of width) holds the title at top-left and three bold-heading-plus-body text blocks stacked down the column, left-aligned at x 3.4%. The right column (about 51 to 100%) is a full-bleed blue-violet photograph with a large rounded dark card centred in it at x 56.8 to 94.5%, y 9.6 to 93.9%. Footer runs along the bottom: OpenAI wordmark at left (x 2.7%), confidentiality line beside it (x 8.2%), page number at bottom right of the photo area (x 96.5%, y 94.4%).

**Visuals**

One diagram, a dark flowchart titled 'Context Building in an LLM Agent' inside a rounded dark card (800 by 1000 px image, with a second 512 by 768 px copy of the same artwork at the same position). Title is top-left in white sans-serif, with an OpenAI knot logo top-right. Under it a legend row reads 'Role:' followed by three green filled circles labelled System, Developer and User. The flow is a vertical chain of shapes: a green-filled rectangle labelled 'Initial prompt' at the top, with a small green leader line from a left label 'Initial prompt' into it; a short downward arrow connects it to a dashed rounded rectangle enclosing a stack of three outlined boxes labelled '"type": "reasoning"', '"type": "function_call"' and '"type": "function_call_output"' (the last wrapped as '"type": "function_' / 'call_output"'); a bracket on the right of the dashed group is labelled 'Full Context'; a downward arrow leads from the stack to a green rectangle 'Next user message' inside the dashed group; a downward arrow leaves the dashed group to an outlined rectangle 'Tool Call Output' with a small green hammer icon at its right end; a thin grey loop line runs down the left side from the Initial prompt area back into the Tool Call Output box; a short downward arrow leads from Tool Call Output to a second green rectangle 'Next user message' at the bottom. Green (#7AC142-ish) for filled role and message boxes, white outlines and white text elsewhere. Composition is a centred vertical portrait, uncropped, with flat even lighting on the dark background.

**Design**

Background: Split background: solid white #FFFFFF on the left half and a blue-violet gradient photograph (#5B5BF0 to #F0EEFF) on the right half. Clean corporate look. Background white #FFFFFF on the left, blue-violet gradient photo on the right fading from #5B5BF0 at left to near-white #F0EEFF top-right. Title colour black #000000; body black #000000 and #1F1F1F for the last block; accent green #7AC142 only inside the diagram. Typeface is OpenAI Sans, a geometric grotesque sans-serif: Medium for the title, Bold for block headings, Regular for body. Title cap height is about 5.9% of slide height (font size 5.93% of height); body text is 3.70% of height; footer 1.48%. Left-aligned throughout. Dark card #1A1A1A with rounded corners of about 3% of its width and a soft drop shadow beneath it; no outlines around text blocks.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 51.3, h 100 [#FFFFFF]: White left-column background rectangle (Covers the left half of the slide; no outline; sits behind all text.)
- photo at x 51.3, y 0, w 48.7, h 100 [#5B5BF0]: Blue-violet gradient background image, full height of right column, no subject (1920 by 1080 px, full-bleed to top, right and bottom edges; gradient from saturated blue-violet #5B5BF0 at left and bottom to pale lavender #F0EEFF at top-right; no overlay.)
- shape at x 56.8, y 9.6, w 37.7, h 84.3 [#1A1A1A]: Dark rounded card behind diagram (Corner radius about 3% of card width; soft dark drop shadow along bottom and right edges; opacity 100%.)
- illustration at x 56.3, y 9, w 38.7, h 86.1 [#1A1A1A]: Context Building in an LLM Agent flowchart (800 by 1000 px image) (Rounded-corner image of the dark diagram described in visuals; sits over the dark card; the dark card edges coincide with the image edges.)
- illustration at x 57.9, y 10.9, w 35.6, h 89.1 [#1A1A1A]: Second copy of the Context Building diagram (512 by 768 px) (Same artwork placed at the same position as the 800 by 1000 image, overlapping it; no separate crop or overlay visible.)
- title at x 2.7, y 5.7, w 30.7, h 13.1 [OpenAI Sans, Medium weight, sans-serif, normal case, no letter-spacing; font size 5.93% of slide height; #000000; left aligned]: What Is Context? / Why does it matter? (Two lines, line spacing about 1.2 times the font size; second line starts at y 12.9%.)
- body at x 3.4, y 28, w 28.9, h 3.7 [OpenAI Sans, Bold weight, sans-serif, normal case; font size 3.7% of slide height; #000000; left aligned]: The Agent’s Working Memory (Block heading.)
- body at x 3.4, y 33.1, w 45.3, h 8.8 [OpenAI Sans, Regular weight, sans-serif, normal case; font size 3.7% of slide height; #000000; left aligned]: Context is the total information an agent has to / complete its task (Two lines, line spacing about 1.4 times the font size.)
- body at x 3.4, y 50.4, w 20.4, h 3.7 [OpenAI Sans, Bold weight, sans-serif, normal case; font size 3.7% of slide height; #000000; left aligned]: The Context Window (Block heading.)
- body at x 3.4, y 57.5, w 40.5, h 8.8 [OpenAI Sans, Regular weight, sans-serif, normal case; font size 3.7% of slide height; #000000; left aligned]: Codex can hold hundreds of documents in / context at once (400-600 pages of text) (Two lines, line spacing about 1.4 times the font size.)
- body at x 3.4, y 72.8, w 14.4, h 3.7 [OpenAI Sans, Bold weight, sans-serif, normal case; font size 3.7% of slide height; #1F1F1F; left aligned]: Why it Matters (Block heading.)
- body at x 3.4, y 77.9, w 44.2, h 8.8 [OpenAI Sans, Regular weight, sans-serif, normal case; font size 3.7% of slide height; #1F1F1F; left aligned]: Focused context leads to safer, more accurate, / and more consistent agent results (Two lines, line spacing about 1.4 times the font size.)
- logo at x 2.7, y 94.3, w 5.9, h 1.6 [#000000]: OpenAI wordmark (Small bold black wordmark in the footer, left of the confidentiality line.)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans, Regular weight, sans-serif, normal case; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.
- page number at x 96.5, y 94.4, w 0.8, h 1.5 [OpenAI Sans, Regular weight, sans-serif, normal case; font size 1.48% of slide height; #000000; right aligned]: 12 (Sits on the blue-violet photo at bottom right.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 56.8, y 9.6, w 37.7, h 84.3: #1A1A1A
- fill at x 3.4, y 72.2, w 14.4, h 4.4: #FFFFFF
- fill at x 3.4, y 77.4, w 44.2, h 4.4: #FFFFFF
- fill at x 3.4, y 82.5, w 32.3, h 4.4: #FFFFFF
- image at x 51.3, y 0.0, w 48.7, h 100.0: 1920 by 1080 pixels (covers the slide)
- image at x 57.9, y 10.9, w 35.6, h 89.1: 512 by 768 pixels
- image at x 56.3, y 9.0, w 38.7, h 86.1: 800 by 1000 pixels
- text at x 2.7, y 5.7, w 26.9, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: What Is Context?
- text at x 2.7, y 12.9, w 30.7, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: Why does it matter?
- text at x 3.4, y 28.0, w 28.9, h 3.7 [OpenAISans-Bold; font size 3.70; #000000]: The Agent’s Working Memory
- text at x 3.4, y 33.1, w 45.3, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Context is the total information an agent has to
- text at x 3.4, y 38.2, w 16.3, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: complete its task
- text at x 3.4, y 50.4, w 20.4, h 3.7 [OpenAISans-Bold; font size 3.70; #000000]: The Context Window
- text at x 3.4, y 57.5, w 40.5, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Codex can hold hundreds of documents in
- text at x 3.4, y 62.6, w 38.4, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: context at once (400-600 pages of text)
- text at x 3.4, y 72.8, w 14.4, h 3.7 [OpenAISans-Bold; font size 3.70; #1F1F1F]: Why it Matters
- text at x 3.4, y 77.9, w 44.2, h 3.7 [OpenAISans-Regular; font size 3.70; #1F1F1F]: Focused context leads to safer, more accurate,
- text at x 3.4, y 83.1, w 32.3, h 3.7 [OpenAISans-Regular; font size 3.70; #1F1F1F]: and more consistent agent results
- text at x 96.5, y 94.4, w 0.8, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: 12
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 13

**Text**

Codex Manages Context / For You
Intelligent Context Compression
Auto compaction runs in the turn loop / when token usage crosses a / model-specific threshold ~90%
OpenAI
Confidential and proprietary.
I'll now zoom into listener-attach and request/response wiring sections in CodexMessageProcessor since those are often prone / to lifecycle issues.
Explored 1 file, 3 searches
I identified several likely review targets; I'll quickly inspect approval/task cleanup and state handling files to confirm whether any / of these map-based state can leak on cancellation or shutdown.
Explored 1 file, 1 search
Context automatically compacted
Worked for 4m 3s
Pulled the latest main and did a broad review of this codebase's major areas (especially the app-server, core, and related / message/turn handling flow), then focused on the largest runtime path ( initialize → message processing → thread/turn / execution → outgoing callbacks ).
1. [P1] Hard panic on unreachable-but-invalid protocol path in CodexMessageProcessor
File: codex_message_processor.rs (lines 488-490)
process_request panics on ClientRequest::Initialize with "Initialize should be handled in / MessageProcessor" .

**Layout**

Two-column split at about 51% of the slide width: a white text column on the left holding the title, bullet list and footer, and a full-bleed gradient panel on the right holding a dark screenshot. The title sits top-left at about 4% margin from the left edge and 4% from the top, in two lines. The bullets are left-aligned in the left column, starting about 27% down the slide. The footer runs along the bottom-left at about 94% of the height, with the OpenAI wordmark followed by the confidentiality line. There is no slide number.

**Visuals**

One photograph-style screenshot of a dark code-assistant transcript window, boxed with rounded corners, placed on the right panel. It shows a dark charcoal (#1A1A1A) chat pane with light grey monospace and sans text: a paragraph about wiring sections in CodexMessageProcessor, a grey line 'Explored 1 file, 3 searches', a paragraph about reviewing approval/task cleanup files, a grey line 'Explored 1 file, 1 search', a divider labelled 'Context automatically compacted' with a small icon, a divider labelled 'Worked for 4m 3s', a summary paragraph with inline code chips (main, app-server, core, initialize, message processing, thread/turn execution, outgoing callbacks), a numbered item '[P1] Hard panic on unreachable-but-invalid protocol path in CodexMessageProcessor', and two bullets: a blue hyperlink 'codex_message_processor.rs (lines 488-490)' and a line about process_request panicking on ClientRequest::Initialize. The text is cut off at the bottom of the window. The window is framed by a thin dark grey edge and a soft shadow, with a faint horizontal bar at the bottom. The right panel behind it is a soft gradient from lavender-white at the top right to saturated blue-violet (#5B6CF2) in the centre and bright blue (#3F6BF0) at the bottom. The screenshot is not cropped and sits fully inside the panel. There are no logos or icons other than the OpenAI wordmark in the footer.

**Design**

Background: Split background: solid white #FFFFFF on the left about 51% of the width, and a blue-violet gradient photograph (lavender-white to #5B6CF2 to #3F6BF0) on the right about 49%. Split layout. The left half is plain white with black titles (#000000) and near-black body text (#1F1F1F). Titles use OpenAI Sans Medium at about 5.9% of slide height. Body bullets use OpenAI Sans Medium at about 3.7% of slide height, with black round bullet dots. The footer uses OpenAI Sans Regular at about 1.5% of slide height. The right panel is a blurred gradient in lavender, violet and blue. The screenshot window has rounded corners of about 1.5% of slide width, a near-black fill (#1A1A1A), and a subtle drop shadow. No strokes are visible on the text. Accents are the blue hyperlink in the screenshot.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: White full-slide background fill (Covers the entire slide; no stroke.)
- photo at x 51.3, y 0, w 48.7, h 100 [#5B6CF2]: Blue-violet gradient panel with soft light blur, full bleed on the right side of the slide (Full height, runs to the right and top and bottom edges; not boxed; no overlay.)
- shape at x 53.9, y 25.5, w 43.6, h 50 [#1A1A1A]: Dark rounded panel behind the screenshot (Corner radius about 1.5% of slide width; soft drop shadow; no stroke.)
- illustration at x 54, y 27.4, w 43.3, h 45.3 [#1A1A1A]: Screenshot of a dark code-assistant transcript window showing Codex output text (see visuals) (Rounded corners matching the panel; not cropped; text in light grey monospace and sans; inline code chips; blue hyperlink.)
- title at x 2.7, y 4.4, w 38.4, h 12.8 [OpenAI Sans Medium, sans-serif, normal weight, no italics, no capitals, normal letter-spacing; font size 5.93% of slide height; #000000; left aligned]: Codex Manages Context / For You (Two lines with line spacing about 1.1; the second line starts at about 11.5% of height.)
- body at x 4.1, y 26.8, w 33.9, h 4.4 [OpenAI Sans Medium, sans-serif, normal weight, no italics, normal letter-spacing; font size 3.7% of slide height; #1F1F1F; left aligned]: Intelligent Context Compression (Bullet list item; round black bullet dot at x 4.1% in ArialMT 3.7% size.)
- body at x 4.1, y 37, w 39.3, h 15 [OpenAI Sans Medium, sans-serif, normal weight, no italics, normal letter-spacing; font size 3.7% of slide height; #1F1F1F; left aligned]: Auto compaction runs in the turn loop / when token usage crosses a / model-specific threshold ~90% (Bullet list item with three wrapped lines at about 5.1% line pitch; round black bullet dot at x 4.1%; text indented to x 7.2%.)
- logo at x 2.7, y 94.3, w 5.6, h 1.6 [#000000; left aligned]: OpenAI wordmark (Black bold wordmark in the bottom-left corner.)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans Regular, sans-serif, normal weight, no italics, normal letter-spacing; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Sits to the right of the OpenAI wordmark on the same baseline.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 4.1, y 26.8, w 33.9, h 4.4: #FFFFFF
- fill at x 4.1, y 37.0, w 39.3, h 4.4: #FFFFFF
- fill at x 7.2, y 42.1, w 26.6, h 4.4: #FFFFFF
- fill at x 7.2, y 47.2, w 29.9, h 4.4: #FFFFFF
- fill at x 53.9, y 25.5, w 43.6, h 50.0: #1A1A1A
- fill at x 54.0, y 27.4, w 43.3, h 45.3: #1A1A1A
- image at x 51.3, y 0.0, w 48.7, h 100.0: 1920 by 1080 pixels (covers the slide)
- image at x 53.4, y 24.8, w 44.6, h 51.8: 500 by 327 pixels
- image at x 53.6, y 26.8, w 44.2, h 46.9: 500 by 299 pixels
- image at x 54.0, y 27.4, w 43.3, h 45.3: 768 by 449 pixels
- text at x 2.7, y 4.4, w 38.4, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: Codex Manages Context
- text at x 2.7, y 11.5, w 11.3, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: For You
- text at x 4.1, y 27.3, w 1.3, h 3.7 [ArialMT; font size 3.70; #1F1F1F]: ●
- text at x 7.2, y 27.3, w 30.9, h 3.7 [OpenAISans-Medium; font size 3.70; #1F1F1F]: Intelligent Context Compression
- text at x 4.1, y 37.6, w 1.3, h 3.7 [ArialMT; font size 3.70; #1F1F1F]: ●
- text at x 7.2, y 37.6, w 36.3, h 3.7 [OpenAISans-Medium; font size 3.70; #1F1F1F]: Auto compaction runs in the turn loop
- text at x 7.2, y 42.7, w 26.6, h 3.7 [OpenAISans-Medium; font size 3.70; #1F1F1F]: when token usage crosses a
- text at x 7.2, y 47.8, w 29.9, h 3.7 [OpenAISans-Medium; font size 3.70; #1F1F1F]: model-speciﬁc threshold ~90%
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 14

**Text**

Q&A
OpenAI

**Layout**

Plain white slide with the title Q&A at top left and the OpenAI wordmark at top right on the same header band. A single full-width image sits below the header, inset with equal left and right margins of about 2.7% of slide width, spanning from about 17% to 100% of slide height minus a small bottom margin. No footer or slide number is visible.

**Visuals**

One photograph-like abstract image in a rounded rectangle, spanning most of the lower slide. Subject: a soft-focus, abstract blur of light and colour, with pale white-lavender and light blue on the upper left fading diagonally into saturated royal blue and violet on the right and bottom right; a faint streaked texture is visible on the left. Composition: landscape crop, full frame inside the rounded box, no visible subject. Lighting: soft, diffuse glow from the upper left. Dominant colours: #E8E6FF, #8FB4FF, #4A7CFF, #3B3BD9, #2E2ED6. The image is boxed, not full-bleed, with small rounded corners and no overlay or border. Next to the Q&A title the OpenAI logo is a black wordmark, no icon.

**Design**

Background: Solid white #FFFFFF. White background, black titles and logo, minimal and airy. Title Q&A in a regular-weight geometric sans-serif (close to Arial/Helvetica Regular), black #000000, about 8.9% of slide height, left aligned. OpenAI wordmark in a bold geometric sans-serif, black #000000, top right. No accent colour apart from the blue image. Rounded corners on the image, about 0.5% of slide height radius. No shadows or gradients on the page itself.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Background fill, solid white (Covers the entire slide, no stroke.)
- title at x 3.6, y 5.8, w 10.5, h 8.9 [ArialMT, regular sans-serif; font size 8.89% of slide height; #000000; left aligned]: Q&A (Single line, no letter-spacing change, regular weight.)
- logo at x 89, y 4.5, w 8.7, h 4 [bold geometric sans-serif; #000000; right aligned]: OpenAI wordmark (Black wordmark logo at top right corner of header band, no icon.)
- photo at x 2.7, y 17, w 94.6, h 83 [#4A7CFF]: Abstract soft-focus blue and lavender gradient photograph in a rounded rectangle, 1920 by 1080 pixels (Rounded corners approx 0.5% of slide height, no border, no shadow, no overlay, not cropped beyond its frame, boxed inside margins.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 2.7, y 17.0, w 94.6, h 83.0: 1920 by 1080 pixels
- text at x 3.6, y 5.8, w 10.5, h 8.9 [ArialMT; font size 8.89; #000000]: Q&A

### Slide 15

**Text**

OpenAI
Using Codex in the
SDLC
Ways to leverage Codex in your work

**Layout**

Two-column split on a white 16:9 slide. The OpenAI logo sits top-left at about x 2.7, y 4.5. The title and subtitle are left-aligned in the left column starting at x 2.3 to 2.8. A rounded-corner image fills the right column from x 24.4 to 100 minus a right margin of about 0.5, with a top margin of 4.8 and bottom margin of about 4.5. No footer or slide number is visible.

**Visuals**

One photograph-style abstract gradient image in the right column, boxed with small rounded corners. It shows soft, blurred, fluid shapes with no recognisable subject. Composition is portrait, full-frame within its box, not cropped to any figure. Lighting is diffuse and glowing. Dominant colours: pale lavender-white at the top-left corner (about #E8E6FB), periwinkle blue (about #6F8BFF) in the centre, saturated royal blue (about #3B5BFF) in the middle-left, and deep indigo-violet (about #3434D8) in the lower-right corner. No overlay or tint beyond the gradient itself. The OpenAI wordmark in black is the only logo.

**Design**

Background: Solid white #FFFFFF. Plain white background (#FFFFFF) with no texture. Title in black (#000000) in a regular-weight Arial-style grotesque sans-serif, about 10.1% of slide height per line, tight leading with the two lines about 12% apart. Subtitle in semibold geometric sans-serif (OpenAI Sans SemiBold), about 4.2% of slide height, in accent blue #3442E0. The logo is black. The image has small rounded corners of about 0.5% of slide width and no shadow or border.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100: Full-slide background fill (Solid fill #FFFFFF covering the whole slide.)
- photo at x 24.4, y 4.8, w 75.6, h 90.7 [#3B5BFF]: Abstract blue and lavender gradient image with soft blurred fluid forms (Placed image 1920 by 1080 pixels, rounded corners about 0.5% of slide width, no border or shadow, not full-bleed; boxed within the right column. Pale lavender at top-left, periwinkle centre, saturated blue middle-left, deep indigo lower-right.)
- logo at x 2.7, y 4.5, w 8.5, h 3.5 [#000000]: OpenAI wordmark (Black OpenAI logotype, top-left corner, sits above the title.)
- title at x 2.3, y 15.5, w 49.7, h 22.3 [ArialMT, regular weight, sans-serif; font size 10.12% of slide height; #000000; left aligned]: Using Codex in the / SDLC (Two lines: 'Using Codex in the' at y 15.5 and 'SDLC' at y 27.7; line spacing about 12.2% of slide height.)
- subtitle at x 2.8, y 41.2, w 40, h 4.2 [OpenAISans-SemiBold, sans-serif, semibold weight; font size 4.2% of slide height; #3442E0; left aligned]: Ways to leverage Codex in your work (Single line in accent blue, sentence case, no letter-spacing change.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 24.4, y 4.8, w 75.6, h 90.7: 1920 by 1080 pixels
- text at x 2.3, y 15.5, w 49.7, h 10.1 [ArialMT; font size 10.12; #000000]: Using Codex in the
- text at x 2.3, y 27.7, w 15.2, h 10.1 [ArialMT; font size 10.12; #000000]: SDLC
- text at x 2.8, y 41.2, w 40.0, h 4.2 [OpenAISans-SemiBold; font size 4.20; #3442E0]: Ways to leverage Codex in your work

### Slide 16

**Review notes**

- Worth studying: A lifecycle table pairs what agents do with what engineers decide at each stage, making the division of labour easy to scan.

**Text**

Codex turbocharges engineers across the entire software lifecycle
Plan
Agents: Fill in missing details and prerequisites
Engineers: Pick priorities and order
Design
Agents: Sketch a starter solution
Engineers: Decide key choices and constraints
Build
Agents: Write code and fix bugs
Engineers: Guide patterns and tradeoffs
Test
Agents: Create and update tests
Engineers: Confirm coverage and intent
Review
Agents: First-pass PR review
Engineers: Final approval and quality bar
Document
Agents: Draft docs and summaries
Engineers: Confirm accuracy and decisions
Deploy & Maintain
Agents: Triage logs and alerts
Engineers: Handle high-risk issues
OpenAI
Confidential and proprietary.

**Layout**

Seven white rounded cards in two staggered rows: four cards across the top row (Plan, Design, Build, Test) and three centred cards below (Review, Document, Deploy & Maintain), joined by short blue connector lines. The title sits left-aligned across the top at about 6% from the top edge with a left margin of about 3.4%; the footer with the OpenAI wordmark and confidentiality line sits bottom-left at about 94.5% from the top with a left margin of about 2.7%. No slide number is shown.

**Visuals**

Seven outline icons, each in the top-right corner of its card, all drawn in black line art with about 0.2% stroke: Plan is a notepad/clipboard with a checkmark; Design is a pen-nib over a circle (drawing tool); Build is a wrench; Test is a magnifying glass; Review is an eye; Document is a pencil; Deploy & Maintain is a speedometer/gauge. Five short blue (#1F1FFF) horizontal lines join adjacent cards: Plan to Design, Design to Build, Build to Test, Review to Document, Document to Deploy & Maintain. The OpenAI wordmark in bold black sits at the bottom left as a logo.

**Design**

Background: Solid white #FFFFFF. Plain white background with black text and black outlines; no accent colours except the blue connector lines. Title in OpenAI Sans Medium, black #000000, about 5.4% of slide height, sentence case. Card headings in OpenAI Sans Bold, black #000000, about 3.7% of slide height. Body text in OpenAI Sans Regular, black #000000, about 2.5% of slide height, set with line spacing of about 1.4 and a blank line between the Agents and Engineers paragraphs. Footer text about 1.5% of slide height. Card outlines are thin black strokes with rounded corners of about 1.2% of slide width, no fill, no shadow, no gradient. Connector lines are blue #1F1FFF, about 0.3% stroke.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 2.7, y 17, w 21.6, h 36.1 [#FFFFFF]: Card outline for Plan (Rounded rectangle, black #000000 stroke about 0.2% of slide width, no shadow)
- shape at x 27, y 17, w 21.6, h 36.1 [#FFFFFF]: Card outline for Design (Rounded rectangle, black #000000 stroke about 0.2% of slide width, no shadow)
- shape at x 51.4, y 17, w 21.6, h 36.1 [#FFFFFF]: Card outline for Build (Rounded rectangle, black #000000 stroke about 0.2% of slide width, no shadow)
- shape at x 75.7, y 17, w 21.6, h 36.1 [#FFFFFF]: Card outline for Test (Rounded rectangle, black #000000 stroke about 0.2% of slide width, no shadow)
- shape at x 16.2, y 55.4, w 21.6, h 36.1 [#FFFFFF]: Card outline for Review (Rounded rectangle, black #000000 stroke about 0.2% of slide width, no shadow)
- shape at x 40.5, y 55.4, w 21.6, h 36.1 [#FFFFFF]: Card outline for Document (Rounded rectangle, black #000000 stroke about 0.2% of slide width, no shadow)
- shape at x 64.9, y 55.4, w 21.6, h 36.1 [#FFFFFF]: Card outline for Deploy & Maintain (Rounded rectangle, black #000000 stroke about 0.2% of slide width, no shadow)
- line at x 24.3, y 34.9, w 2.7, h 0.3 [#1F1FFF]: Connector from Plan to Design (Horizontal blue line about 0.3% stroke, at vertical centre of cards)
- line at x 48.6, y 34.9, w 2.7, h 0.3 [#1F1FFF]: Connector from Design to Build (Horizontal blue line about 0.3% stroke)
- line at x 73, y 34.7, w 2.7, h 0.3 [#1F1FFF]: Connector from Build to Test (Horizontal blue line about 0.3% stroke)
- line at x 37.8, y 73.3, w 2.7, h 0.3 [#1F1FFF]: Connector from Review to Document (Horizontal blue line about 0.3% stroke)
- line at x 62.1, y 73.3, w 2.7, h 0.3 [#1F1FFF]: Connector from Document to Deploy & Maintain (Horizontal blue line about 0.3% stroke)
- icon at x 20.3, y 18.5, w 2.7, h 4.6 [#000000]: Notepad with checkmark icon (Plan) (Black line icon, about 0.2% stroke, top-right of Plan card)
- icon at x 45.1, y 18.3, w 2.7, h 5 [#000000]: Pen nib over circle icon (Design) (Black line icon, top-right of Design card)
- icon at x 69.4, y 18.3, w 2.9, h 5 [#000000]: Wrench icon (Build) (Black line icon, top-right of Build card)
- icon at x 93.8, y 18.4, w 2.8, h 4.9 [#000000]: Magnifying glass icon (Test) (Black line icon, top-right of Test card)
- icon at x 33.5, y 56.7, w 3.2, h 4.6 [#000000]: Eye icon (Review) (Black line icon, top-right of Review card)
- icon at x 58.8, y 56.7, w 2.6, h 4.6 [#000000]: Pencil icon (Document) (Black line icon, top-right of Document card)
- icon at x 82.9, y 56.9, w 2.9, h 4.1 [#000000]: Speedometer/gauge icon (Deploy & Maintain) (Black line icon, top-right of Deploy & Maintain card)
- title at x 3.4, y 5.8, w 92.3, h 5.4 [OpenAI Sans Medium, regular weight, sentence case; font size 5.43% of slide height; #000000; left aligned]: Codex turbocharges engineers across the entire software lifecycle
- body at x 4.5, y 22.3, w 4.3, h 3.7 [OpenAI Sans Bold, sentence case; font size 3.7% of slide height; #000000; left aligned]: Plan
- body at x 4.5, y 29, w 15.5, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Agents: Fill in missing / details and prerequisites (Line spacing about 1.4; blank line before next paragraph)
- body at x 4.5, y 39.2, w 16, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Engineers: Pick priorities / and order (Line spacing about 1.4)
- body at x 28.7, y 22.3, w 6.7, h 3.7 [OpenAI Sans Bold, sentence case; font size 3.7% of slide height; #000000; left aligned]: Design
- body at x 28.7, y 29, w 15.6, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Agents: Sketch a starter / solution (Line spacing about 1.4)
- body at x 28.7, y 39.2, w 15, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Engineers: Decide key / choices and constraints (Line spacing about 1.4)
- body at x 53.1, y 22.3, w 5, h 3.7 [OpenAI Sans Bold, sentence case; font size 3.7% of slide height; #000000; left aligned]: Build
- body at x 53.1, y 29, w 17, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Agents: Write code and fix / bugs (Line spacing about 1.4)
- body at x 53.1, y 39.2, w 16.8, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Engineers: Guide patterns / and tradeoffs (Line spacing about 1.4)
- body at x 77.4, y 22.3, w 4.1, h 3.7 [OpenAI Sans Bold, sentence case; font size 3.7% of slide height; #000000; left aligned]: Test
- body at x 77.4, y 29, w 17.4, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Agents: Create and update / tests (Line spacing about 1.4)
- body at x 77.4, y 39.2, w 12.6, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Engineers: Confirm / coverage and intent (Line spacing about 1.4)
- body at x 18, y 62, w 7.1, h 3.7 [OpenAI Sans Bold, sentence case; font size 3.7% of slide height; #000000; left aligned]: Review
- body at x 18, y 68.7, w 14, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Agents: First-pass PR / review (Line spacing about 1.4)
- body at x 18, y 78.9, w 16.3, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Engineers: Final approval / and quality bar (Line spacing about 1.4)
- body at x 42.2, y 62, w 10.2, h 3.7 [OpenAI Sans Bold, sentence case; font size 3.7% of slide height; #000000; left aligned]: Document
- body at x 42.2, y 68.7, w 15, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Agents: Draft docs and / summaries (Line spacing about 1.4)
- body at x 42.2, y 78.9, w 14.7, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Engineers: Confirm / accuracy and decisions (Line spacing about 1.4)
- body at x 66.6, y 62, w 17.9, h 3.7 [OpenAI Sans Bold, sentence case; font size 3.7% of slide height; #000000; left aligned]: Deploy & Maintain
- body at x 66.6, y 68.7, w 18.7, h 2.5 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Agents: Triage logs and alerts
- body at x 66.6, y 75.5, w 17.7, h 5.9 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Engineers: Handle high-risk / issues (Line spacing about 1.4)
- logo at x 2.7, y 94.5, w 3, h 1.3 [#000000]: OpenAI wordmark (Bold black wordmark, bottom-left corner)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans Regular; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 2.7, y 17.0, w 21.6, h 36.1: #FFFFFF
- fill at x 27.0, y 17.0, w 21.6, h 36.1: #FFFFFF
- fill at x 51.4, y 17.0, w 21.6, h 36.1: #FFFFFF
- fill at x 75.7, y 17.0, w 21.6, h 36.1: #FFFFFF
- fill at x 16.2, y 55.4, w 21.6, h 36.1: #FFFFFF
- fill at x 40.5, y 55.4, w 21.6, h 36.1: #FFFFFF
- fill at x 64.9, y 55.4, w 21.6, h 36.1: #FFFFFF
- text at x 3.4, y 5.8, w 92.3, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: Codex turbocharges engineers across the entire software lifecycle
- text at x 4.5, y 22.3, w 4.3, h 3.7 [OpenAISans-Bold; font size 3.70; #000000]: Plan
- text at x 28.7, y 22.3, w 6.7, h 3.7 [OpenAISans-Bold; font size 3.70; #000000]: Design
- text at x 53.1, y 22.3, w 5.0, h 3.7 [OpenAISans-Bold; font size 3.70; #000000]: Build
- text at x 77.4, y 22.3, w 4.1, h 3.7 [OpenAISans-Bold; font size 3.70; #000000]: Test
- text at x 4.5, y 29.0, w 14.1, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Agents: Fill in missing
- text at x 28.7, y 29.0, w 15.6, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Agents: Sketch a starter
- text at x 53.1, y 29.0, w 17.0, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Agents: Write code and ﬁx
- text at x 77.4, y 29.0, w 17.4, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Agents: Create and update
- text at x 4.5, y 32.4, w 15.5, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: details and prerequisites
- text at x 28.7, y 32.4, w 5.0, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: solution
- text at x 53.1, y 32.4, w 3.1, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: bugs
- text at x 77.4, y 32.4, w 3.1, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: tests
- text at x 4.5, y 39.2, w 16.0, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Engineers: Pick priorities
- text at x 28.7, y 39.2, w 14.3, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Engineers: Decide key
- text at x 53.1, y 39.2, w 16.8, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Engineers: Guide patterns
- text at x 77.4, y 39.2, w 12.6, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Engineers: Conﬁrm
- text at x 18.0, y 62.0, w 7.1, h 3.7 [OpenAISans-Bold; font size 3.70; #000000]: Review
- text at x 42.2, y 62.0, w 10.2, h 3.7 [OpenAISans-Bold; font size 3.70; #000000]: Document
- text at x 66.6, y 62.0, w 17.9, h 3.7 [OpenAISans-Bold; font size 3.70; #000000]: Deploy & Maintain
- text at x 4.5, y 42.6, w 6.1, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: and order
- text at x 28.7, y 42.6, w 15.0, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: choices and constraints
- text at x 53.1, y 42.6, w 8.5, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: and tradeoff s
- text at x 77.4, y 42.6, w 12.6, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: coverage and intent
- text at x 18.0, y 68.7, w 14.0, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Agents: First-pass PR
- text at x 42.2, y 68.7, w 15.0, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Agents: Draft docs and
- text at x 66.6, y 68.7, w 18.7, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Agents: Triage logs and alerts
- text at x 18.0, y 72.1, w 4.1, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: review
- text at x 42.2, y 72.1, w 7.0, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: summaries
- text at x 66.6, y 75.5, w 17.7, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Engineers: Handle high-risk
- text at x 18.0, y 78.9, w 16.3, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Engineers: Final approval
- text at x 42.2, y 78.9, w 12.6, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Engineers: Conﬁrm
- text at x 66.6, y 78.9, w 3.9, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: issues
- text at x 18.0, y 82.3, w 9.5, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: and quality bar
- text at x 42.2, y 82.3, w 14.7, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: accuracy and decisions
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 17

**Text**

Time to build together
Send a read-only prompt with @ mention
Build a simple feature (+ run tests)
Build a harder feature using “plan mode”
Add image context
Review code before deploying
Document changes
Let’s build
codex
Please review my changes and develop a effective unit testing strategy for these changes.
Please list out all of the edge cases that need to be tested along with anything we may have missed
GPT-5.3-Codex
High
IDE context
Local
Custom (config.toml)
daniels-cai/skill-duplicatio...

**Layout**

Two-column split slide: the left white column (about 0 to 51% of width) holds the title and a bulleted list; the right side (about 51 to 100% of width) is a full-height blue-violet gradient panel with a dark rounded app window inset from it. The title sits top-left at about 6.8% from the top with a left margin of about 3.4%; the bullets are left-aligned with a left margin of about 4%, spaced roughly 9.3% of slide height apart, starting at about 25.6% from the top. No footer, no slide number is visible.

**Visuals**

Screenshot of a dark application window (Codex IDE) shown inside the right panel. The window is a near-black (#1A1A1A) rounded rectangle with a thin light-grey edge and soft shadow, spanning about 57% to 92% of slide width and 4% to 96% of slide height. At the top centre is a white outline cloud-shaped logo containing a terminal prompt symbol '>_' (about 0.5 to 0.7 of slide height in size). Below it, centred, the white text 'Let's build' with 'codex' in grey beneath it with a small down chevron. A rounded input box near the bottom holds the prompt text in white, and beneath it a toolbar with a plus icon, a 'GPT-5.3-Codex' dropdown, 'High' dropdown, a blue 'IDE context' label with a sparkle icon, a microphone icon and a white circular up-arrow send button. Below that is a status bar with 'Local' dropdown, 'Custom (config.toml)' dropdown with a gear icon, and a git branch 'daniels-cai/skill-duplicatio...' dropdown. The screenshot is full, not cropped, boxed as a rounded window on a gradient background. The gradient runs from saturated blue #4B5BEA at left to lighter violet and near-white #F2F0FF at top-right and bottom-right #3D6BF0 blue; overlay none.

**Design**

Background: Split background: left half solid white #FFFFFF; right half blue-violet gradient from #4B5BEA to #F2F0FF. Clean minimal sans-serif look. Left panel background pure white #FFFFFF; title in black #000000 in a geometric sans-serif (OpenAI Sans Regular), about 5.9% of slide height, regular weight. Bullets in black #000000, OpenAI Sans Regular, about 3.7% of slide height, with black round bullet glyphs (ArialMT). Right panel uses a blue-violet gradient background (#4B5BEA to #F2F0FF to #3D6BF0). App window is #1A1A1A with rounded corners of about 2% radius and a subtle shadow; screenshot text is light grey-white #F0F0F0 in a sans-serif, with the 'IDE context' label in blue #5B9BFF.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: White full-slide background fill (Covers entire slide; left half visible.)
- photo at x 51.3, y 0, w 48.7, h 100 [#4B5BEA]: Blue-violet gradient panel (background image) covering right half of slide with soft lighter top-right corner (Full-height image, 1920 by 1080 pixels, placed to the right half, not boxed.)
- shape at x 54.9, y 3, w 39.2, h 93.9 [#1A1A1A]: Dark rounded app window frame (Corner radius about 2% of slide width, thin light-grey edge, soft drop shadow.)
- illustration at x 54.4, y 2.4, w 40.2, h 95.7 [#1A1A1A]: Screenshot of Codex app window with 'Let's build codex' start screen, prompt input and status bar (Image 747 by 1000 pixels, overlaps the dark window frame.)
- logo at x 57.2, y 4.2, w 34.5, h 91.5 [#FFFFFF]: White outline cloud logo with terminal prompt '>_' at top centre of app window (Placed screenshot image 424 by 632 pixels spanning the window region.)
- title at x 3.4, y 6.8, w 33.8, h 5.9 [OpenAI Sans, Regular, sans-serif; font size 5.93% of slide height; #000000; left aligned]: Time to build together
- body at x 4, y 25.6, w 1.3, h 3.7 [Arial, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: ● Send a read-only prompt with @ mention (Bullet glyph, black round bullet.)
- body at x 6.5, y 25.6, w 39.5, h 3.7 [OpenAI Sans, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: Send a read-only prompt with @ mention
- body at x 4, y 35, w 1.3, h 3.7 [Arial, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: ● Build a simple feature (+ run tests) (Bullet glyph, black round bullet.)
- body at x 6.5, y 35, w 32.6, h 3.7 [OpenAI Sans, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: Build a simple feature (+ run tests)
- body at x 4, y 44.3, w 1.3, h 3.7 [Arial, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: ● Build a harder feature using “plan mode” (Bullet glyph, black round bullet.)
- body at x 6.5, y 44.3, w 38.5, h 3.7 [OpenAI Sans, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: Build a harder feature using “plan mode”
- body at x 4, y 53.6, w 1.3, h 3.7 [Arial, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: ● Add image context (Bullet glyph, black round bullet.)
- body at x 6.5, y 53.6, w 18, h 3.7 [OpenAI Sans, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: Add image context
- body at x 4, y 63, w 1.3, h 3.7 [Arial, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: ● Review code before deploying (Bullet glyph, black round bullet.)
- body at x 6.5, y 63, w 28.5, h 3.7 [OpenAI Sans, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: Review code before deploying
- body at x 4, y 72.3, w 1.3, h 3.7 [Arial, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: ● Document changes (Bullet glyph, black round bullet.)
- body at x 6.5, y 72.3, w 18.5, h 3.7 [OpenAI Sans, Regular, sans-serif; font size 3.7% of slide height; #000000; left aligned]: Document changes

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 54.9, y 3.0, w 39.2, h 93.9: #1A1A1A
- image at x 51.3, y 0.0, w 48.7, h 100.0: 1920 by 1080 pixels (covers the slide)
- image at x 54.4, y 2.4, w 40.2, h 95.7: 747 by 1000 pixels
- image at x 57.2, y 4.2, w 34.5, h 91.5: 424 by 632 pixels
- text at x 3.4, y 6.8, w 33.8, h 5.9 [OpenAISans-Regular; font size 5.93; #000000]: Time to build together
- text at x 4.0, y 25.6, w 1.3, h 3.7 [ArialMT; font size 3.70; #000000]: ●
- text at x 6.5, y 25.6, w 39.5, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Send a read-only prompt with @ mention
- text at x 4.0, y 35.0, w 1.3, h 3.7 [ArialMT; font size 3.70; #000000]: ●
- text at x 6.5, y 35.0, w 32.6, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Build a simple feature (+ run tests)
- text at x 4.0, y 44.3, w 1.3, h 3.7 [ArialMT; font size 3.70; #000000]: ●
- text at x 6.5, y 44.3, w 38.5, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Build a harder feature using “plan mode”
- text at x 4.0, y 53.6, w 1.3, h 3.7 [ArialMT; font size 3.70; #000000]: ●
- text at x 6.5, y 53.6, w 18.0, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Add image context
- text at x 4.0, y 63.0, w 1.3, h 3.7 [ArialMT; font size 3.70; #000000]: ●
- text at x 6.5, y 63.0, w 28.5, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Review code before deploying
- text at x 4.0, y 72.3, w 1.3, h 3.7 [ArialMT; font size 3.70; #000000]: ●
- text at x 6.5, y 72.3, w 18.5, h 3.7 [OpenAISans-Regular; font size 3.70; #000000]: Document changes

### Slide 18

**Text**

Delegate to codex, and free yourself to think
more deeply about architecture and design
OpenAI
Confidential and proprietary.
18

**Layout**

A single centred white rounded card sits in the middle of a full-bleed blue background, spanning about 14% to 86% horizontally and 18% to 82% vertically. Inside the card a two-line centred headline sits above a centred app icon, with nothing else in the card. The footer runs along the bottom with the OpenAI wordmark at left, the confidentiality line beside it, and a tiny page number at the bottom right.

**Visuals**

One photographic-style background image covering the full slide (1920 by 1080 px), a soft blurred gradient of saturated blue and violet on the left and bottom, fading to pale lavender and white at the top centre and right with faint streaky highlights. One app icon, 200 by 200 px, centred in the card at about x 43.8%, y 48.4%, 12.6% by 22.4% of the slide: a rounded-square tile with a near-white grey (#F3F3F5) glossy surface and a thin grey outline, containing a cloud-shaped badge with a violet-to-blue gradient (top #A88BF0 to bottom #3D2EF5) and a glossy highlight, with a white chevron '>' and a white horizontal dash '_' inside it, reading as a terminal prompt. The icon is not cropped and has no visible drop shadow.

**Design**

Background is a blurred blue-violet gradient photograph with white highlights. The card is solid white #FFFFFF with large rounded corners and no visible border or shadow. Headline text is black #000000 in OpenAI Sans, a geometric grotesque sans-serif, with Bold for 'Delegate' and 'free yourself' and Regular for the rest; line size about 4.4% of slide height, sentence case, no italics or letter-spacing. Footer text is black #000000 in OpenAI Sans Regular about 1.5% of slide height, and the page number is black in Arial, about 0.6%. Accents are the violet-blue of the icon; there are no rules or extra shapes.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- photo at x 0.1, y 0, w 99.9, h 100: Full-bleed blurred blue-violet gradient background image with pale lavender and white streaks, 1920 by 1080 px (Covers the whole slide, not cropped into a box, no overlay; dominant colour #4F5BF0 with pale #E6E8FA highlights)
- shape at x 14.3, y 18.1, w 71.4, h 63.8 [#FFFFFF]: White rounded rectangle card (Corner radius about 3% of slide height, no stroke, no shadow, opacity 100%, sits over the background image)
- icon at x 43.8, y 48.4, w 12.6, h 22.4 [#F3F3F5]: App icon: glossy violet-to-blue cloud badge with white terminal prompt chevron and dash, inside a near-white rounded-square tile (Image 200 by 200 px, tile has thin grey outline and rounded corners, cloud fill gradient #A88BF0 to #3D2EF5, white chevron and dash strokes, not cropped, no shadow, sits on the white card)
- title at x 25.8, y 33.2, w 51.5, h 9.7 [OpenAI Sans, Regular with Bold for 'Delegate' and 'free yourself', sans-serif, no italics, no capitals, no letter-spacing; font size 4.44% of slide height; #000000; center aligned]: Delegate to codex, and free yourself to think / more deeply about architecture and design (Two lines, first line 'Delegate to codex, and free yourself to think' and second line 'more deeply about architecture and design', line spacing about 1.3 times the font size)
- logo at x 2.7, y 94.4, w 5.6, h 1.5 [#000000]: OpenAI wordmark (Bold black wordmark in the bottom-left corner, estimated position)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans, Regular, sans-serif, no italics, no capitals, no letter-spacing; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.
- page number at x 96, y 94.8, w 0.4, h 0.6 [Arial, Regular, sans-serif; font size 0.62% of slide height; #000000; left aligned]: 18

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 14.3, y 18.1, w 71.4, h 63.8: #FFFFFF
- image at x 0.1, y 0.0, w 99.9, h 100.0: 1920 by 1080 pixels (covers the slide)
- image at x 43.8, y 48.4, w 12.6, h 22.4: 200 by 200 pixels
- text at x 25.8, y 33.2, w 10.5, h 4.4 [OpenAISans-Bold; font size 4.44; #000000]: Delegate
- text at x 36.3, y 33.2, w 16.5, h 4.4 [OpenAISans-Regular; font size 4.44; #000000]: to codex, and
- text at x 52.8, y 33.2, w 14.7, h 4.4 [OpenAISans-Bold; font size 4.44; #000000]: free yourself
- text at x 67.5, y 33.2, w 9.8, h 4.4 [OpenAISans-Regular; font size 4.44; #000000]: to think
- text at x 26.8, y 38.5, w 49.0, h 4.4 [OpenAISans-Regular; font size 4.44; #000000]: more deeply about architecture and design
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.
- text at x 96.0, y 94.8, w 0.4, h 0.6 [ArialMT; font size 0.62; #000000]: 18

### Slide 19

**Text**

Q&A
OpenAI

**Layout**

Plain white slide with the title 'Q&A' at top left and the 'OpenAI' wordmark at top right, both sitting in the top band. A single full-width image is boxed below the header, inset with equal side margins of about 2.7% and a bottom margin of about 3%. No footer or slide number is visible.

**Visuals**

One photograph-like abstract image in a rounded rectangular box spanning the content width below the header. It shows a soft, blurred close-up of a glossy surface with diagonal streaks: pale white and lavender at upper left, shading into bright cornflower blue across the centre, and deep indigo-violet at lower right and upper right. Lighting is soft and diffuse with a bright highlight near the top centre-right edge. Composition is landscape, full-width, unobstructed, with no overlay or tint. The image has small corner radii and is not cropped to a circle. No logos or icons other than the OpenAI wordmark.

**Design**

Background: Solid white #FFFFFF. White background #FFFFFF. Title 'Q&A' in a light-weight geometric sans-serif (closest to Helvetica/Arial regular) in black #000000, roughly 8.9% of slide height for cap height-scaled size. OpenAI wordmark in a bold geometric sans-serif, black #000000. Image has rounded corners of about 0.5% of slide height and no shadow. No accent colours beyond the image palette of white, lavender #E6E3FF, cornflower blue #5B8CFF, and indigo #3A36D8.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide white background fill (Covers the entire slide, no border.)
- photo at x 2.7, y 17, w 94.6, h 83 [#5B8CFF]: Abstract blurred blue and violet gradient photograph with diagonal light streaks, full-width boxed image (Rounded rectangle crop with corner radius about 0.5% of slide height, no border, no shadow, no overlay; dominant colours white-lavender #E6E3FF at upper left, cornflower #5B8CFF centre, indigo #3A36D8 lower right.)
- title at x 3.6, y 5.8, w 10.5, h 8.9 [ArialMT, regular weight; font size 8.89% of slide height; #000000; left aligned]: Q&A (Single line, no italics, no capitals transform, default letter-spacing.)
- logo at x 89, y 4.5, w 8, h 3.5 [bold geometric sans-serif; #000000; right aligned]: OpenAI wordmark (Top-right corner logo, black text wordmark with no icon.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 2.7, y 17.0, w 94.6, h 83.0: 1920 by 1080 pixels
- text at x 3.6, y 5.8, w 10.5, h 8.9 [ArialMT; font size 8.89; #000000]: Q&A

### Slide 20

**Text**

OpenAI / How To Effectively / Delegate to Agents / Become a power user fast / Codex wrote 85% / of the projects’ code

**Layout**

Two-column layout on a white slide: the text column sits on the left and a large rounded-corner blue gradient panel fills the right side, running nearly to the top and bottom edges. The OpenAI wordmark sits top-left at the margin, the title is left-aligned beneath it with the subtitle below, and the caption is centred horizontally inside the blue panel under the white card. No footer, page number or source line is visible.

**Visuals**

One illustration in the right panel: a white rounded-corner card (about 1.5% of slide width corner radius) holding a 6-column by 5-row grid of square tiles with rounded corners. Each tile is either a light-blue tile (#A9C8F5 to #B3D0F7) with a purple-to-magenta terminal-prompt glyph (a small cloud-like shape containing a '>_' prompt symbol) or a plain light-green tile (#C6F0C2) with a thin darker-green border. Row 1 reads blue, green, blue, green, blue, green. Row 2 is all blue. Row 3 reads blue, blue, green, blue, green, blue. Row 4 reads blue, blue, green, blue, green, blue. Row 5 reads blue, green, blue, green, blue, green. That gives 20 blue tiles and 10 green tiles. The grid is evenly spaced with generous white padding inside the card. The blue panel behind it is a soft gradient from pale lavender #E9E7FB at upper left through violet #7B7CF0 to royal blue #2F3FD8 at lower right, with a slightly lighter blue haze at mid-left, and no photograph content. The panel is flat-lit with no visible light source, is not cropped inside the slide, and has no overlay beyond the gradient. The OpenAI logo is a black wordmark, not a separate icon.

**Design**

Background: Solid white #FFFFFF behind the full slide, with the blue gradient image panel placed on the right side. White background #FFFFFF with a blue gradient panel on the right. Title text is black #000000 in a regular-weight neo-grotesque sans-serif resembling Arial (ArialMT), set large at about 10.1% of slide height per line with tight leading. Subtitle is semibold #3442E0 in OpenAI Sans-style sans-serif at about 4.2% of slide height. Caption is white #FFFFFF in OpenAI Sans Regular at about 4.4% of slide height with '85%' in Bold. The blue panel has small corner radius (about 0.5% of slide width). The white card has a larger corner radius (about 1.5% of slide width) and no visible shadow. No borders on the card. Tiles have rounded corners of about 0.4% of slide width.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide white background fill covering the entire slide (Solid fill, no border, no shadow, opacity 100%.)
- photo at x 24.4, y 4.8, w 75.6, h 90.7 [#2F3FD8]: Blue-to-lavender gradient panel image (1920 by 1080 px placed image) whose visible content is the rounded blue gradient panel on the right; the remaining area of the placed image box is white (Visible panel spans approximately x 56.8 to 97.3 and y 4.8 to 95.6 with rounded corners of about 0.5% of slide width. Gradient runs from pale lavender #E9E7FB upper left through violet #7B7CF0 to royal blue #2F3FD8 lower right. No overlay or tint beyond the gradient. Not full-bleed.)
- illustration at x 60.6, y 24.5, w 33.9, h 40.2 [#FFFFFF]: White rounded card containing a 6-by-5 grid of tiles: 20 light-blue tiles with purple terminal-prompt glyphs and 10 plain light-green tiles, arranged as described in visuals (Card fill #FFFFFF with corner radius about 1.5% of slide width, no border, no shadow, sits on top of the blue panel. Tile fills: blue #A9C8F5, green #C6F0C2 with darker green border #5DAA55; glyph purple-magenta #9B4FD8 on blue tiles. Tiles evenly spaced with rounded corners.)
- logo at x 2.7, y 4.6, w 7.9, h 3.9 [#000000; left aligned]: OpenAI wordmark (Black wordmark in bold geometric sans, top-left corner logo.)
- title at x 2.3, y 15.5, w 48.1, h 22.3 [ArialMT, regular, sentence-case, letter-spacing normal; font size 10.12% of slide height; #000000; left aligned]: How To Effectively / Delegate to Agents (Two lines, first line box x 2.3 y 15.5 w 47.9 h 10.1; second line box x 2.3 y 27.7 w 48.1 h 10.1. Line spacing about 12.2% of slide height between baselines.)
- subtitle at x 2.8, y 41.2, w 28.2, h 4.2 [OpenAISans-SemiBold, sans-serif, normal case, letter-spacing normal; font size 4.2% of slide height; #3442E0; left aligned]: Become a power user fast (Single line, royal-blue text.)
- caption at x 65.9, y 66.8, w 23.2, h 9.8 [OpenAISans-Regular, sans-serif, normal case; '85%' in OpenAISans-Bold; letter-spacing normal; font size 4.44% of slide height; #FFFFFF; center aligned]: Codex wrote 85% / of the projects’ code (Two centred lines; line 1 box x 67.4 y 66.8 w 14.9 h 4.4 for 'Codex wrote' plus x 82.3 w 5.3 for '85%'; line 2 box x 65.9 y 72.2 w 23.2 h 4.4. Sits on the blue panel below the white card. Line spacing about 5.4% of slide height.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 24.4, y 4.8, w 75.6, h 90.7: 1920 by 1080 pixels
- image at x 60.6, y 24.5, w 33.9, h 40.2: 384 by 256 pixels
- text at x 2.3, y 15.5, w 47.9, h 10.1 [ArialMT; font size 10.12; #000000]: How To Effectively
- text at x 2.3, y 27.7, w 48.1, h 10.1 [ArialMT; font size 10.12; #000000]: Delegate to Agents
- text at x 2.8, y 41.2, w 28.2, h 4.2 [OpenAISans-SemiBold; font size 4.20; #3442E0]: Become a power user fast
- text at x 67.4, y 66.8, w 14.9, h 4.4 [OpenAISans-Regular; font size 4.44; #FFFFFF]: Codex wrote
- text at x 82.3, y 66.8, w 5.3, h 4.4 [OpenAISans-Bold; font size 4.44; #FFFFFF]: 85%
- text at x 65.9, y 72.2, w 23.2, h 4.4 [OpenAISans-Regular; font size 4.44; #FFFFFF]: of the projects’ code

### Slide 21

**Text**

What power users know
01 Instead of a perfect prompt, build perfect context
02 Faster validation means better outcomes
03 Use the agent to learn, not vibe code
OpenAI
Confidential and proprietary.
21

**Layout**

Two-column layout on a white field: a left title column starting at about 4.4% from the left edge with the two-line title top-aligned near the top, and a right column of three numbered statements starting at about 29.7% from the left, spaced down the slide at roughly 7%, 29% and 52% of the height. The lower-left area is empty. The footer sits at the bottom with the OpenAI wordmark at left and the confidentiality line beside it, and the slide number sits at the bottom right.

**Visuals**

none

**Design**

Background: Solid white #FFFFFF. Plain white background with no gradient, shadow or ornament. Title and body text are black #000000 in a geometric sans-serif (the title appears to be Arial-like regular weight; the numbered statements are OpenAI Sans medium with semibold numerals). Title text is about 5.9% of slide height (cap height about 4.2%), numbered statements about 4.4% of slide height, footer text about 1.5% and slide number about 0.6%. No accent colours, no lines, no shapes; hierarchy is set by size and weight alone.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide white background rectangle (Fills the whole slide, no border, no shadow, opacity 100%.)
- title at x 4.4, y 6.1, w 17.8, h 13 [ArialMT, regular weight, sentence case; font size 5.93% of slide height; #000000; left aligned]: What power / users know (Two lines with line break after 'power'; line spacing about 7.1% of slide height between baselines.)
- body at x 29.7, y 7, w 2.6, h 4.4 [OpenAI Sans, semibold weight, numerals; font size 4.44% of slide height; #000000; left aligned]: 01 (Numbered label of first statement.)
- body at x 35.4, y 7, w 56.7, h 4.4 [OpenAI Sans, medium weight, sentence case; font size 4.44% of slide height; #000000; left aligned]: Instead of a perfect prompt, build perfect context (Single line, no bullets.)
- body at x 29.7, y 29.1, w 3.1, h 4.4 [OpenAI Sans, semibold weight, numerals; font size 4.44% of slide height; #000000; left aligned]: 02 (Numbered label of second statement.)
- body at x 35.4, y 29.1, w 47, h 4.4 [OpenAI Sans, medium weight, sentence case; font size 4.44% of slide height; #000000; left aligned]: Faster validation means better outcomes (Single line, no bullets.)
- body at x 29.7, y 51.8, w 3.1, h 4.4 [OpenAI Sans, semibold weight, numerals; font size 4.44% of slide height; #000000; left aligned]: 03 (Numbered label of third statement.)
- body at x 35.4, y 51.8, w 41.9, h 4.4 [OpenAI Sans, medium weight, sentence case; font size 4.44% of slide height; #000000; left aligned]: Use the agent to learn, not vibe code (Single line, no bullets.)
- logo at x 2.7, y 94.3, w 5.3, h 1.6 [#000000]: OpenAI wordmark (Black OpenAI wordmark at bottom left, sits on the same baseline as the footer text.)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [Helvetica Neue, regular weight, sentence case; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.
- page number at x 96, y 94.8, w 0.4, h 0.6 [Arial regular; font size 0.62% of slide height; #000000; right aligned]: 21

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- text at x 4.4, y 6.1, w 17.8, h 5.9 [ArialMT; font size 5.93; #000000]: What power
- text at x 29.7, y 7.0, w 2.6, h 4.4 [OpenAISans-SemiBold; font size 4.44; #000000]: 01
- text at x 35.4, y 7.0, w 56.7, h 4.4 [OpenAISans-Medium; font size 4.44; #000000]: Instead of a perfect prompt, build perfect context
- text at x 4.4, y 13.2, w 16.9, h 5.9 [ArialMT; font size 5.93; #000000]: users know
- text at x 29.7, y 29.1, w 3.1, h 4.4 [OpenAISans-SemiBold; font size 4.44; #000000]: 02
- text at x 35.4, y 29.1, w 47.0, h 4.4 [OpenAISans-Medium; font size 4.44; #000000]: Faster validation means better outcomes
- text at x 29.7, y 51.8, w 3.1, h 4.4 [OpenAISans-SemiBold; font size 4.44; #000000]: 03
- text at x 35.4, y 51.8, w 41.9, h 4.4 [OpenAISans-Medium; font size 4.44; #000000]: Use the agent to learn, not vibe code
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.
- text at x 96.0, y 94.8, w 0.4, h 0.6 [ArialMT; font size 0.62; #000000]: 21

### Slide 22

**Review notes**

- Worth studying: A four-field template with short headings (Goal, Context Pointers, Constraints, Done When) gives a clear, reusable structure for the argument.

**Text**

Mini Codex prompt template
Four fields that keep work scoped,
reviewable, and repeatable
Goal
The outcome you want, what
you are changing, and why it
matters
Context Pointers
Relevant files or any
examples of desired
behavior or existing patterns
Constraints
Do and do not rules,
conventions, and style
expectations
Done When
Tests and checks to pass,
commands to run, and
expected behavior verified
OpenAI
Confidential and proprietary.

**Layout**

Single slide with a title block top-left and a row of four equal-width boxes spanning the content width, all aligned to the same top edge, with a small footer at bottom-left. Side margins are about 2.7% of slide width, the title starts at x 4.4% and y 6.1%, and the boxes sit at y 35.6% to 75.0%. There is no slide number.

**Visuals**

Four identical rounded-corner outlined rectangles in a horizontal row, each with an icon above a heading and a short description. Thin black connector lines run horizontally between adjacent boxes at mid-height (y about 55.3%), joining box 1 to box 2, box 2 to box 3, and box 3 to box 4. Box 1 icon: a line icon of four small open circles in a 2x2 grid with a plus sign in the bottom-right position. Box 2 icon: a globe line icon with a vertical ellipse and horizontal and vertical meridian lines. Box 3 icon: a line icon of a robot head with a rectangular face showing two dot eyes above a rounded body outline. Box 4 icon: a code icon showing a left chevron, a forward slash, and a right chevron. All icons are solid black line glyphs of similar stroke weight. No photographs or charts.

**Design**

Background: Solid white #FFFFFF. Plain white background with no texture or gradient. Title and body text are black #000000; headings are black #000000; icons are black #000000; boxes have a black #000000 outline of about 1.5 px at 2000 px width with no fill (white #FFFFFF interior) and rounded corners with a radius of about 0.7% of slide width; connector lines are black #000000 about 1.5 px thick. No shadows. Title is an ArialMT regular sans-serif at about 5.9% of slide height, subtitle is ArialMT regular at about 3.7%, card headings are Helvetica Neue Bold sans-serif at about 3.95%, body text is Helvetica Neue regular sans-serif at about 2.47%, footer text is Helvetica Neue regular at about 1.48%, and the OpenAI wordmark is a small bold sans-serif. Line spacing in body text is about 3.4% of slide height.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 2.7, y 35.6, w 21.6, h 39.4 [#FFFFFF]: Rounded rectangle box 1 (outline only, no text) (Fill #FFFFFF, stroke black #000000 about 1.5 px, corner radius about 0.7% of slide width, no shadow)
- shape at x 27.1, y 35.6, w 21.6, h 39.4 [#FFFFFF]: Rounded rectangle box 2 (outline only, no text) (Fill #FFFFFF, stroke black #000000 about 1.5 px, corner radius about 0.7% of slide width, no shadow)
- shape at x 51.4, y 35.6, w 21.6, h 39.4 [#FFFFFF]: Rounded rectangle box 3 (outline only, no text) (Fill #FFFFFF, stroke black #000000 about 1.5 px, corner radius about 0.7% of slide width, no shadow)
- shape at x 75.8, y 35.6, w 21.6, h 39.4 [#FFFFFF]: Rounded rectangle box 4 (outline only, no text) (Fill #FFFFFF, stroke black #000000 about 1.5 px, corner radius about 0.7% of slide width, no shadow)
- line at x 24.3, y 55.3, w 2.7, h 0 [#000000]: Connector line between box 1 and box 2 (Horizontal black stroke about 1.5 px, joins right edge of box 1 to left edge of box 2 at mid-height)
- line at x 48.7, y 55.3, w 2.7, h 0 [#000000]: Connector line between box 2 and box 3 (Horizontal black stroke about 1.5 px, joins right edge of box 2 to left edge of box 3 at mid-height)
- line at x 73, y 55.3, w 2.7, h 0 [#000000]: Connector line between box 3 and box 4 (Horizontal black stroke about 1.5 px, joins right edge of box 3 to left edge of box 4 at mid-height)
- title at x 4.4, y 6.1, w 41.7, h 5.9 [ArialMT, regular, sans-serif, no italics, no capitals, normal letter-spacing; font size 5.93% of slide height; #000000; left aligned]: Mini Codex prompt template
- subtitle at x 4.4, y 12.8, w 32.3, h 7.6 [ArialMT, regular, sans-serif, no italics, no capitals, normal letter-spacing; font size 3.7% of slide height; #000000; left aligned]: Four fields that keep work scoped, / reviewable, and repeatable (Two lines, line pitch about 4.5% of slide height)
- icon at x 4.9, y 40.4, w 2.9, h 5.1 [#000000]: Icon for Goal: four open circles in a 2x2 grid with a plus sign at bottom-right (Black line glyph, stroke about 2 px, sits at top-left of box 1)
- heading at x 4.6, y 51.2, w 4.9, h 4 [HelveticaNeue-Bold, bold, sans-serif, no italics, no capitals, normal letter-spacing; font size 3.95% of slide height; #000000; left aligned]: Goal
- body at x 4.6, y 58.1, w 18.4, h 9 [HelveticaNeue, regular, sans-serif, no italics, no capitals, normal letter-spacing; font size 2.47% of slide height; #000000; left aligned]: The outcome you want, what / you are changing, and why it / matters (Line pitch about 3.4% of slide height, no bullets)
- icon at x 29, y 39.7, w 2.8, h 5.1 [#000000]: Icon for Context Pointers: globe line icon with vertical ellipse and horizontal and vertical meridian lines (Black line glyph, stroke about 2 px, sits at top-left of box 2)
- heading at x 28.8, y 51.2, w 17.7, h 4 [HelveticaNeue-Bold, bold, sans-serif, no italics, no capitals, normal letter-spacing; font size 3.95% of slide height; #000000; left aligned]: Context Pointers
- body at x 28.8, y 58.1, w 17.6, h 9 [HelveticaNeue, regular, sans-serif, no italics, no capitals, normal letter-spacing; font size 2.47% of slide height; #000000; left aligned]: Relevant files or any / examples of desired / behavior or existing patterns (Line pitch about 3.4% of slide height, no bullets)
- icon at x 53.7, y 39.7, w 2.5, h 4.7 [#000000]: Icon for Constraints: robot head line icon with rectangular face, two dot eyes and rounded body outline (Black line glyph, stroke about 2 px, sits at top-left of box 3)
- heading at x 53.2, y 51.2, w 12.3, h 4 [HelveticaNeue-Bold, bold, sans-serif, no italics, no capitals, normal letter-spacing; font size 3.95% of slide height; #000000; left aligned]: Constraints
- body at x 53.2, y 58.1, w 14.3, h 9 [HelveticaNeue, regular, sans-serif, no italics, no capitals, normal letter-spacing; font size 2.47% of slide height; #000000; left aligned]: Do and do not rules, / conventions, and style / expectations (Line pitch about 3.4% of slide height, no bullets)
- icon at x 78, y 40.7, w 3.3, h 3.3 [#000000]: Icon for Done When: code line icon showing left chevron, forward slash and right chevron (Black line glyph, stroke about 2 px, sits at top-left of box 4)
- heading at x 77.5, y 51.2, w 12.2, h 4 [HelveticaNeue-Bold, bold, sans-serif, no italics, no capitals, normal letter-spacing; font size 3.95% of slide height; #000000; left aligned]: Done When
- body at x 77.5, y 58.1, w 16.5, h 9 [HelveticaNeue, regular, sans-serif, no italics, no capitals, normal letter-spacing; font size 2.47% of slide height; #000000; left aligned]: Tests and checks to pass, / commands to run, and / expected behavior verified (Line pitch about 3.4% of slide height, no bullets)
- logo at x 2.7, y 94.4, w 2.9, h 1.4 [bold sans-serif wordmark; font size 1.5% of slide height; #000000; left aligned]: OpenAI wordmark (Bottom-left corner logo)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue, regular, sans-serif, no italics, no capitals, normal letter-spacing; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Sits to the right of the OpenAI wordmark on the same baseline)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 2.7, y 35.6, w 21.6, h 39.4: #FFFFFF
- fill at x 27.1, y 35.6, w 21.6, h 39.4: #FFFFFF
- fill at x 51.4, y 35.6, w 21.6, h 39.4: #FFFFFF
- fill at x 75.8, y 35.6, w 21.6, h 39.4: #FFFFFF
- text at x 4.4, y 6.1, w 41.7, h 5.9 [ArialMT; font size 5.93; #000000]: Mini Codex prompt template
- text at x 4.4, y 12.8, w 32.3, h 3.7 [ArialMT; font size 3.70; #000000]: Four fields that keep work scoped,
- text at x 4.4, y 17.3, w 25.0, h 3.7 [ArialMT; font size 3.70; #000000]: reviewable, and repeatable
- text at x 4.6, y 51.2, w 4.9, h 4.0 [HelveticaNeue-Bold; font size 3.95; #000000]: Goal
- text at x 28.8, y 51.2, w 17.7, h 4.0 [HelveticaNeue-Bold; font size 3.95; #000000]: Context Pointers
- text at x 53.2, y 51.2, w 12.3, h 4.0 [HelveticaNeue-Bold; font size 3.95; #000000]: Constraints
- text at x 77.5, y 51.2, w 12.2, h 4.0 [HelveticaNeue-Bold; font size 3.95; #000000]: Done When
- text at x 4.6, y 58.1, w 18.4, h 2.5 [HelveticaNeue; font size 2.47; #000000]: The outcome you want, what
- text at x 28.8, y 58.1, w 12.9, h 2.5 [HelveticaNeue; font size 2.47; #000000]: Relevant ﬁles or any
- text at x 53.2, y 58.1, w 13.0, h 2.5 [HelveticaNeue; font size 2.47; #000000]: Do and do not rules,
- text at x 77.5, y 58.1, w 16.5, h 2.5 [HelveticaNeue; font size 2.47; #000000]: Tests and checks to pass,
- text at x 4.6, y 61.5, w 18.2, h 2.5 [HelveticaNeue; font size 2.47; #000000]: you are changing, and why it
- text at x 28.8, y 61.5, w 12.9, h 2.5 [HelveticaNeue; font size 2.47; #000000]: examples of desired
- text at x 53.2, y 61.5, w 14.3, h 2.5 [HelveticaNeue; font size 2.47; #000000]: conventions, and style
- text at x 77.5, y 61.5, w 14.4, h 2.5 [HelveticaNeue; font size 2.47; #000000]: commands to run, and
- text at x 4.6, y 65.0, w 4.7, h 2.5 [HelveticaNeue; font size 2.47; #000000]: matters
- text at x 28.8, y 65.0, w 17.6, h 2.5 [HelveticaNeue; font size 2.47; #000000]: behavior or existing patterns
- text at x 53.2, y 65.0, w 7.9, h 2.5 [HelveticaNeue; font size 2.47; #000000]: expectations
- text at x 77.5, y 65.0, w 16.4, h 2.5 [HelveticaNeue; font size 2.47; #000000]: expected behavior veriﬁed
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 23

**Review notes**

- What would change the score: Merge or shrink the four repeated Q&A pages, and add a short stated takeaway to the dense text slides 23 and 27 so the audience sees the point rather than reading paragraphs.

**Text**

Prompting: Why it matters + how to do it well
A good prompt sets scope, constraints, and success criteria so Codex can execute predictably
Keep it minimal: goal + constraints + “done”: Codex performs best with a clear target and a finish line, not a long narrative. Tell Codex what not to do when needed
Paste the full stack trace when debugging: Include the error, repro steps, and environment details so Codex can pinpoint root cause faster
Bake in verification: Specify how you will validate (tests, lint, build). If it cannot run them, it should tell you exactly what to run and what to look for
Use open ended prompts intentionally: Ask for options, tradeoffs, risks, and recommended approaches when you want guidance, not implementation
Provide clear code pointers with @ mention: Reference the relevant files in the codebase to keep edits scoped and accurate
Resource: developers.openai.com/codex/prompting/
23
Confidential and proprietary.
OpenAI

**Layout**

Title block is top-left at about 6.6% from the top with a one-line subtitle directly beneath it, both left-aligned at x 3.3%. Below is a two-column body: the left column starts at x 3.8% and the right column at x 51.9%, with the right column's first block starting lower at y 34.0%, leaving an empty band at the top right. The resource line is centred near y 84.6%, and a footer row sits at y 94.4% with the OpenAI logo at left, the confidentiality note beside it, and the slide number at the right margin.

**Visuals**

none

**Design**

Background: Solid white #FFFFFF. Plain white background with black sans-serif text throughout. Title is OpenAI Sans Medium at about 5.9% of slide height; subtitle and body are OpenAI Sans Medium at about 3.7% and 3.2%; bold lead-ins use OpenAI Sans Bold at the body size. Footer and resource line are Helvetica Neue at about 1.5% and 3.2%. The resource URL is underlined. No shapes, rules, shadows, gradients or corner radii.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide background fill (Solid fill covering the whole slide, no border.)
- title at x 3.3, y 6.6, w 68.7, h 5.9 [OpenAI Sans, Medium, sans-serif; font size 5.93% of slide height; #000000; left aligned]: Prompting: Why it matters + how to do it well
- subtitle at x 3.3, y 13.4, w 89.4, h 3.7 [OpenAI Sans, Medium, sans-serif; font size 3.7% of slide height; #000000; left aligned]: A good prompt sets scope, constraints, and success criteria so Codex can execute predictably
- body at x 3.8, y 23.9, w 44.4, h 15.9 [OpenAI Sans, Medium with Bold lead-in, sans-serif; font size 3.21% of slide height; #000000; left aligned]: Keep it minimal: goal + constraints + “done”: Codex / performs best with a clear target and a finish line, not / a long narrative. Tell Codex what not to do when / needed (Lead-in ‘Keep it minimal: goal + constraints + “done”:’ in OpenAI Sans Bold; remainder Medium. Line spacing about 4.2% of slide height.)
- body at x 51.9, y 34, w 42.9, h 11.6 [OpenAI Sans, Medium with Bold lead-in, sans-serif; font size 3.21% of slide height; #000000; left aligned]: Paste the full stack trace when debugging: Include / the error, repro steps, and environment details so / Codex can pinpoint root cause faster (Lead-in ‘Paste the full stack trace when debugging:’ in OpenAI Sans Bold; remainder Medium. Line spacing about 4.2% of slide height.)
- body at x 3.8, y 47.8, w 43, h 11.6 [OpenAI Sans, Medium with Bold lead-in, sans-serif; font size 3.21% of slide height; #000000; left aligned]: Bake in verification: Specify how you will validate / (tests, lint, build). If it cannot run them, it should tell / you exactly what to run and what to look for (Lead-in ‘Bake in verification:’ in OpenAI Sans Bold; remainder Medium. Line spacing about 4.2% of slide height.)
- body at x 51.9, y 53.6, w 39.3, h 15.9 [OpenAI Sans, Medium with Bold lead-in, sans-serif; font size 3.21% of slide height; #000000; left aligned]: Use open ended prompts intentionally: Ask for / options, tradeoffs, risks, and recommended / approaches when you want guidance, not / implementation (Lead-in ‘Use open ended prompts intentionally:’ in OpenAI Sans Bold; remainder Medium. Line spacing about 4.2% of slide height.)
- body at x 3.8, y 67.4, w 43.2, h 11.7 [OpenAI Sans, Medium with Bold lead-in, sans-serif; font size 3.21% of slide height; #000000; left aligned]: Provide clear code pointers with @ mention: / Reference the relevant files in the codebase to keep / edits scoped and accurate (Lead-in ‘Provide clear code pointers with @ mention:’ in OpenAI Sans Bold; remainder Medium. Line spacing about 4.2% of slide height.)
- caption at x 28.7, y 84.6, w 42.6, h 3.2 [Helvetica Neue, Regular, sans-serif; font size 3.21% of slide height; #000000; center aligned]: Resource: developers.openai.com/codex/prompting/ (‘Resource:’ in regular weight; the URL developers.openai.com/codex/prompting/ is a hyperlink underlined in the same black.)
- logo at x 2.7, y 94.4, w 5.5, h 1.5 [#000000]: OpenAI wordmark (Small bold black wordmark, text logo, bottom-left corner.)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [Helvetica Neue, Regular, sans-serif; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.
- page number at x 96.4, y 94.4, w 0.9, h 1.5 [Helvetica Neue, Regular, sans-serif; font size 1.48% of slide height; #000000; right aligned]: 23

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- text at x 3.3, y 6.6, w 68.7, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: Prompting: Why it matters + how to do it well
- text at x 3.3, y 13.4, w 89.4, h 3.7 [OpenAISans-Medium; font size 3.70; #000000]: A good prompt sets scope, constraints, and success criteria so Codex can execute predictably
- text at x 3.8, y 23.9, w 37.3, h 3.2 [OpenAISans-Bold; font size 3.21; #000000]: Keep it minimal: goal + constraints + “done”:
- text at x 41.1, y 23.9, w 6.1, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: Codex
- text at x 3.8, y 28.2, w 44.4, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: performs best with a clear target and a ﬁnish line, not
- text at x 3.8, y 32.4, w 39.9, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: a long narrative. Tell Codex what not to do when
- text at x 51.9, y 34.0, w 36.0, h 3.2 [OpenAISans-Bold; font size 3.21; #000000]: Paste the full stack trace when debugging:
- text at x 87.9, y 34.0, w 6.9, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: Include
- text at x 3.8, y 36.6, w 6.2, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: needed
- text at x 51.9, y 38.2, w 40.9, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: the error, repro steps, and environment details so
- text at x 51.9, y 42.4, w 30.5, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: Codex can pinpoint root cause faster
- text at x 3.8, y 47.8, w 17.1, h 3.2 [OpenAISans-Bold; font size 3.21; #000000]: Bake in veriﬁcation:
- text at x 21.0, y 47.8, w 24.1, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: Specify how you will validate
- text at x 3.8, y 52.0, w 42.9, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: (tests, lint, build). If it cannot run them, it should tell
- text at x 51.9, y 53.6, w 33.1, h 3.2 [OpenAISans-Bold; font size 3.21; #000000]: Use open ended prompts intentionally:
- text at x 85.0, y 53.6, w 6.2, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: Ask for
- text at x 3.8, y 56.2, w 36.1, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: you exactly what to run and what to look for
- text at x 51.9, y 57.8, w 36.2, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: options, tradeoff s, risks, and recommended
- text at x 51.9, y 62.0, w 34.6, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: approaches when you want guidance, not
- text at x 51.9, y 66.3, w 13.1, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: implementation
- text at x 3.8, y 67.4, w 37.2, h 3.2 [OpenAISans-Bold; font size 3.21; #000000]: Provide clear code pointers with @ mention:
- text at x 3.8, y 71.6, w 43.2, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: Reference the relevant ﬁles in the codebase to keep
- text at x 3.8, y 75.9, w 21.8, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: edits scoped and accurate
- text at x 28.7, y 84.6, w 42.6, h 3.2 [HelveticaNeue; font size 3.21; #000000]: Resource: developers.openai.com/codex/prompting/
- text at x 96.4, y 94.4, w 0.9, h 1.5 [HelveticaNeue; font size 1.48; #000000]: 23
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 24

**Text**

Adding AGENTS.md helps make Codex smarter
Codex starts each run with limited project knowledge
AGENTS.md makes your conventions and workflows load automatically
Result: more consistent edits, fewer retries, safer changes
Use /init to generate a starter AGENTS.md
AGENTS.md
A simple, open format for guiding coding agents.
Codex from OpenAI
Amp
Jules from Google
Cursor
Factory
RooCode

**Layout**

Single-column title across the top left at about x 3.4%, y 5.9%, with no footer or slide number. The body is a left column of four bulleted points starting at x 4.8% (bullet) and x 7.9% (text), y 32.8% to 67.3%. A single illustration sits in the right column from x 47.9% to 98.1%, y 23.3% to 77.2%, vertically centred against the bullets' block.

**Visuals**

One illustration, a rounded-corner screenshot-style graphic in the right column. Subject: a presentation card titled 'AGENTS.md' with the subtitle 'A simple, open format for guiding coding agents.' and a 3-by-2 grid of tool logos with labels: row 1 Codex (from OpenAI, terminal-prompt icon in a dark outlined circle), Amp (dark stacked chevron-stripe mark), Jules (from Google, dark octopus-like mark); row 2 Cursor (grey faceted 3D cube), Factory (dark asterisk/flower-like mark), RooCode (dark kangaroo silhouette). Composition: the white card is centred inside a larger sky-blue frame, with the card occupying about 55% to 90% of the frame width and 32% to 68% of its height. Lighting: flat, even. Dominant colours: sky blue #4DC8FF frame, white card #FFFFFF, yellow-green #C8D92E and violet #9B6BE0 blurred glow blobs, icons dark slate #2B3445. The frame is not cropped, has no border, and has a soft drop shadow beneath it. Overlay: faint white monospace code fragments (for example 'eal wi', 'hape in', 'tensorfl', 'k e', 'ax') are scattered over the blue frame edges at low opacity. The white card is a rounded rectangle with a faint soft shadow.

**Design**

Background: Solid white #FFFFFF. Plain white slide with black sans-serif headline and near-black body text. Title is a geometric sans-serif (OpenAI Sans-like), Medium weight, about 5.9% of slide height, black #000000. Body bullets are the same family in Medium weight at about 3.5% of slide height, colour #030A18; bullet glyphs are small round dots in the same colour. Accent colours appear only in the illustration. Line spacing in the bullets is about 1.4 times the font size, with about 1.6 lines of space between bullets. Shapes: the illustration frame has rounded corners of about 2% of slide width and a soft grey drop shadow; the inner card has rounded corners and a faint shadow. No gradient on the slide background.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100: Full-slide background fill (Solid fill #FFFFFF covering the whole slide, no border.)
- illustration at x 47.9, y 23.3, w 50.2, h 53.9: Sky-blue rounded frame with blurred yellow-green and violet glow and faint monospace code fragments, containing the AGENTS.md card (Dominant fill #4DC8FF, rounded corners about 2% of slide width, soft drop shadow below, not cropped, not full-bleed, boxed on white.)
- illustration at x 50.2, y 27.2, w 45.5, h 45.5: White card titled AGENTS.md with subtitle 'A simple, open format for guiding coding agents.' and six tool logos in a 3-by-2 grid: Codex from OpenAI, Amp, Jules from Google, Cursor, Factory, RooCode (Card fill #FFFFFF with faint soft shadow and rounded corners; 16:9 image; title 'AGENTS.md' bold dark slate #2B3445; subtitle regular dark slate; logos dark slate #2B3445 with small grey labels; overlaps the blue frame; not cropped.)
- title at x 3.4, y 5.9, w 72.6, h 5.9 [OpenAISans-Medium; font size 5.93% of slide height; #000000; left aligned]: Adding AGENTS.md helps make Codex smarter (Sans-serif Medium, single line, no capitals transform.)
- body at x 4.8, y 32.8, w 38, h 34.5 [OpenAISans-Medium; font size 3.46% of slide height; #030A18; left aligned]: • Codex starts each run with limited project / knowledge / • AGENTS.md makes your conventions and / workflows load automatically / • Result: more consistent edits, fewer retries, / safer changes / • Use /init to generate a starter AGENTS.md (Bulleted list, small round dot bullets at x 4.8%, hanging text indent at x 7.9%; paragraph gaps between bullets; line breaks as shown.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 47.9, y 23.3, w 50.2, h 53.9: 1000 by 604 pixels
- image at x 50.2, y 27.2, w 45.5, h 45.5: 512 by 288 pixels
- text at x 3.4, y 5.9, w 72.6, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: Adding AGENTS.md helps make Codex smarter
- text at x 4.8, y 32.8, w 0.6, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: •
- text at x 7.9, y 32.8, w 37.7, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: Codex starts each run with limited project
- text at x 7.9, y 37.6, w 9.7, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: knowledge
- text at x 4.8, y 44.3, w 0.6, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: •
- text at x 7.9, y 44.3, w 37.6, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: AGENTS.md makes your conventions and
- text at x 7.9, y 49.1, w 26.1, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: workﬂows load automatically
- text at x 4.8, y 55.8, w 0.6, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: •
- text at x 7.9, y 55.8, w 38.9, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: Result: more consistent edits, fewer retries,
- text at x 7.9, y 60.6, w 12.5, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: safer changes
- text at x 4.8, y 67.3, w 0.6, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: •
- text at x 7.9, y 67.3, w 37.8, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: Use /init to generate a starter AGENTS.md

### Slide 25

**Text**

How Codex discovers guidance
Discovery follows this precedence order
Global (~/.codex/AGENTS.md)
• Codex uses only the first non-empty file at the
home directory level
• Keep concise and use for personal defaults
Project (AGENTS.md)
Directory (AGENTS.md)
• Starts at the project root (usually Git root) and
walks down to the current working directory
• Concatenates files from the root down
• Files closer to current directory override earlier
guidance for specific context and guidance

**Layout**

Two-column layout on a white field. The title and subtitle sit top-left, left-aligned at about 3.4% from the left edge. The left column holds a boxed file-tree screenshot starting at about 29.8% from the top. The right column, starting at about 54% from the left, holds a globe-headed Global section, a horizontal rule at about 44% height, and a folder-headed Project section with a nested Directory sub-heading and bullet list. There is no footer, no page number and no logo.

**Visuals**

One image: a monospace file-tree screenshot, about 50.7% of slide width by 44.7% of slide height, placed at x 1.6%, y 29.8%. It shows a directory tree with a book-style file icon beside each file. Top row: AGENTS.md with the annotation 'Repository expectations', in a rounded light-green bordered bar. Below it, services/ (chevron expanded) containing payments/ (expanded) with AGENTS.md annotated 'Ignored because an override exists' (plain), AGENTS.override.md in a rounded light-green bordered bar annotated 'Payments service rules', and README.md (plain). Then search/ (expanded) containing AGENTS.md (plain) and a generic file icon followed by '…'. Green highlights (#2E9E4F text, #F3F9F3 fill, #8CC98F border) mark the two active files. The tree is set on a white background with grey tree connector lines. Icons: a globe outline icon (black, about 2.6% wide) beside the Global heading; a folder outline icon (black, about 2.6% wide) beside the Project heading; a folder outline icon (black, about 2.6% wide) beside the Directory heading. A thin grey horizontal rule (#8A8A8A, about 0.2% high) spans the right column at about 44% height.

**Design**

Background: Solid white #FFFFFF. Clean white corporate style with a single dark ink colour for all text. Title is OpenAI Sans Regular in black #000000 at about 5.9% of slide height, which is large. Subtitle is OpenAI Sans Regular in dark grey #282828 at about 3.7% of slide height. Section headings are OpenAI Sans Bold in near-black navy #030A18 at about 3.95% of slide height. Bullet text is OpenAI Sans Regular in #030A18 at about 3.46% of slide height, with small round bullets and single line spacing. Accent colour is green (#2E9E4F) used only inside the screenshot. Shape styles: the tree highlight bars have rounded corners and a thin green stroke; the divider is a plain hairline. No shadows, gradients or corner radii elsewhere.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide white background fill (Covers the entire slide, no stroke.)
- title at x 3.4, y 5.9, w 47.4, h 5.9 [OpenAI Sans Regular; font size 5.93% of slide height; #000000; left aligned]: How Codex discovers guidance
- subtitle at x 3.4, y 12.6, w 37.8, h 3.7 [OpenAI Sans Regular; font size 3.7% of slide height; #282828; left aligned]: Discovery follows this precedence order
- photo at x 1.6, y 29.8, w 50.7, h 44.7 [#FFFFFF]: File-tree screenshot showing AGENTS.md discovery in services/payments and services/search directories with annotations (Image 737 by 365 pixels, placed unboxed on white with no crop. Monospace tree text with green highlighted rounded bars for AGENTS.md and AGENTS.override.md.)
- icon at x 54.5, y 17.5, w 2.6, h 4 [#030A18]: Globe icon (Black outline globe with meridian and equator lines.)
- subtitle at x 58.6, y 17.7, w 32, h 4 [OpenAI Sans Bold; font size 3.95% of slide height; #030A18; left aligned]: Global (~/.codex/AGENTS.md)
- body at x 54.2, y 25.4, w 43.1, h 3.5 [OpenAI Sans Regular; font size 3.46% of slide height; #030A18; left aligned]: • Codex uses only the first non-empty file at the / home directory level (Bullet marker is a small round dot. Line spacing about 4.8% of slide height between wrapped lines.)
- body at x 54.2, y 35, w 39.8, h 3.5 [OpenAI Sans Regular; font size 3.46% of slide height; #030A18; left aligned]: • Keep concise and use for personal defaults (Bullet marker is a small round dot.)
- line at x 53.8, y 44.3, w 43.8, h 0.2 [#8A8A8A]: Horizontal divider rule (Hairline grey rule about 1 pixel thick spanning the right column.)
- icon at x 54.5, y 49.5, w 2.6, h 4 [#030A18]: Folder icon for Project (Black outline folder.)
- subtitle at x 58.7, y 50, w 23, h 4 [OpenAI Sans Bold; font size 3.95% of slide height; #030A18; left aligned]: Project (AGENTS.md)
- icon at x 58.5, y 57.5, w 2.6, h 4 [#030A18]: Folder icon for Directory (Black outline folder, indented under Project.)
- subtitle at x 63, y 57.4, w 25.2, h 4 [OpenAI Sans Bold; font size 3.95% of slide height; #030A18; left aligned]: Directory (AGENTS.md)
- body at x 54.2, y 67.1, w 43.6, h 19.3 [OpenAI Sans Regular; font size 3.46% of slide height; #030A18; left aligned]: • Starts at the project root (usually Git root) and / walks down to the current working directory / • Concatenates files from the root down / • Files closer to current directory override earlier / guidance for specific context and guidance (Four bulleted items with round dot markers; wrapped lines are unbulleted continuations. Each text line is a separate measured text box at y 67.1, 71.9, 76.6, 81.4 and 86.2.)
- caption at x 0, y 0, w 0, h 0: Slide number and footer: none (No page number, footer or source line is printed on this slide.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 3.4, y 12.0, w 37.8, h 4.4: #FFFFFF
- image at x 1.6, y 29.8, w 50.7, h 44.7: 737 by 365 pixels
- text at x 3.4, y 5.9, w 47.4, h 5.9 [OpenAISans-Regular; font size 5.93; #000000]: How Codex discovers guidance
- text at x 3.4, y 12.6, w 37.8, h 3.7 [OpenAISans-Regular; font size 3.70; #282828]: Discovery follows this precedence order
- text at x 58.6, y 17.7, w 32.0, h 4.0 [OpenAISans-Bold; font size 3.95; #030A18]: Global (~/.codex/AGENTS.md)
- text at x 54.2, y 25.4, w 43.1, h 3.5 [OpenAISans-Regular; font size 3.46; #030A18]: • Codex uses only the ﬁrst non-empty ﬁle at the
- text at x 56.1, y 30.2, w 18.0, h 3.5 [OpenAISans-Regular; font size 3.46; #030A18]: home directory level
- text at x 54.2, y 35.0, w 39.8, h 3.5 [OpenAISans-Regular; font size 3.46; #030A18]: • Keep concise and use for personal defaults
- text at x 58.7, y 50.0, w 23.0, h 4.0 [OpenAISans-Bold; font size 3.95; #030A18]: Project (AGENTS.md)
- text at x 63.0, y 57.4, w 25.2, h 4.0 [OpenAISans-Bold; font size 3.95; #030A18]: Directory (AGENTS.md)
- text at x 54.2, y 67.1, w 43.3, h 3.5 [OpenAISans-Regular; font size 3.46; #030A18]: • Starts at the project root (usually Git root) and
- text at x 56.1, y 71.9, w 38.9, h 3.5 [OpenAISans-Regular; font size 3.46; #030A18]: walks down to the current working directory
- text at x 54.2, y 76.6, w 35.9, h 3.5 [OpenAISans-Regular; font size 3.46; #030A18]: • Concatenates ﬁles from the root down
- text at x 54.2, y 81.4, w 43.6, h 3.5 [OpenAISans-Regular; font size 3.46; #030A18]: • Files closer to current directory override earlier
- text at x 56.1, y 86.2, w 38.2, h 3.5 [OpenAISans-Regular; font size 3.46; #030A18]: guidance for speciﬁc context and guidance

### Slide 26

**Text**

What AGENTS.md typically includes
Common sections:
● Repo map: key directories, important modules
● How to run the project: install, build, test, lint
● Engineering guardrails: conventions, PR expectations,
“do not” rules, security constraints
● How work gets verified: what “done” means and how
to validate it
Tip: You can always ask Codex to do a retrospective and
update the Agents.md for you to avoid future mistakes or
get to an end result faster

**Layout**

Single-slide two-column layout on white: title top-left spanning the upper band, a left text column of bulleted list and tip paragraph, and a screenshot panel at the right that is cropped off the right edge of the slide. No footer or slide number is visible.

**Visuals**

One screenshot of a code document (sample AGENTS.md file) in a white rounded panel with a thin light grey border and soft shadow, placed at the right and cropped by the right slide edge. The monospace text reads: '# Sample AGENTS.md file' / '## Dev environment tips' / '- Use `pnpm dlx turbo run where <project_name>` to jump to a package instead' / '- Run `pnpm install --filter <project_name>` to add the package to your works' / '- Use `pnpm create vite@latest <project_name> -- --template react-ts` to spin' / '- Check the name field inside each package's package.json to confirm the right' / '## Testing instructions' / '- Find the CI plan in the .github/workflows folder.' / '- Run `pnpm turbo run test --filter <project_name>` to run every check define' / '- From the package root you can just call `pnpm test`. The commit should pass' / '- To focus on one step, add the Vitest pattern: `pnpm vitest run -t "<test na' / '- Fix any test or type errors until the whole suite is green.' / '- After moving files or changing imports, run `pnpm lint --filter <project_na' / '- Add or update tests for the code you change, even if nobody asked.' / '## PR instructions' / '- Title format: [<project_name>] <Title>' / '- Always run `pnpm lint` and `pnpm test` before committing.' Inline code spans have light grey-blue highlight chips. Dominant colours: black monospace text on white and pale grey-blue chips. The image is cropped at the right edge of the slide and at the right edge of the panel.

**Design**

Background: Solid white #FFFFFF covering the full slide. White background (#FFFFFF). Title in OpenAI Sans Medium, black #000000, about 5.9% of slide height for the cap-height box (roughly 7% em size). Body text in OpenAI Sans Medium, near-black navy #030A18, about 3.5% of slide height. Bullets are filled black circles (ArialMT glyph) in #030A18. Accent colour: none apart from the grey screenshot panel. Screenshot panel has a white fill, light grey border and soft shadow with small corner radius. Sans-serif geometric typefaces throughout; code image uses a monospace font (similar to Courier-like or IBM Plex Mono style). Generous left margin about 3.5% of width.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: White full-slide background fill (Covers the entire slide, no border.)
- title at x 3.4, y 5.9, w 55.4, h 5.9 [OpenAI Sans, Medium weight, normal case, no italics, no letter-spacing; font size 5.93% of slide height; #000000; left aligned]: What AGENTS.md typically includes
- photo at x 57.2, y 15.6, w 42.8, h 68.8 [#FFFFFF]: Screenshot of sample AGENTS.md code document (monospace text in white rounded panel, cropped at right slide edge) (Image 1024 by 463 pixels, white panel with thin light grey border and soft shadow, small corner radius, cropped by right slide edge; inline code chips in pale grey-blue (#E6EAF0 approx).)
- body at x 3.7, y 19.6, w 16.7, h 3.5 [OpenAI Sans, Regular/Medium weight, normal case; font size 3.46% of slide height; #030A18; left aligned]: Common sections:
- body at x 4.5, y 26.4, w 1.2, h 3.5 [Arial, regular; font size 3.46% of slide height; #030A18; left aligned]: ● (Filled circle bullet.)
- body at x 7.5, y 26.4, w 41.1, h 3.5 [OpenAI Sans, Medium weight, normal case; font size 3.46% of slide height; #030A18; left aligned]: Repo map: key directories, important modules
- body at x 4.5, y 31.1, w 1.2, h 3.5 [Arial, regular; font size 3.46% of slide height; #030A18; left aligned]: ● (Filled circle bullet.)
- body at x 7.5, y 31.1, w 39.7, h 3.5 [OpenAI Sans, Medium weight, normal case; font size 3.46% of slide height; #030A18; left aligned]: How to run the project: install, build, test, lint
- body at x 4.5, y 35.9, w 1.2, h 3.5 [Arial, regular; font size 3.46% of slide height; #030A18; left aligned]: ● (Filled circle bullet.)
- body at x 7.5, y 35.9, w 48.4, h 18 [OpenAI Sans, Medium weight, normal case; font size 3.46% of slide height; #030A18; left aligned]: Engineering guardrails: conventions, PR expectations, / “do not” rules, security constraints (Two lines with wrapped second line starting at x 7.5, y 40.7.)
- body at x 4.5, y 45.4, w 1.2, h 3.5 [Arial, regular; font size 3.46% of slide height; #030A18; left aligned]: ● (Filled circle bullet.)
- body at x 7.5, y 45.4, w 47.2, h 8.3 [OpenAI Sans, Medium weight, normal case; font size 3.46% of slide height; #030A18; left aligned]: How work gets verified: what “done” means and how / to validate it (Two lines; second line starts at x 7.5, y 50.2.)
- body at x 3.7, y 63.7, w 51.2, h 12.9 [OpenAI Sans, Medium weight, normal case; font size 3.46% of slide height; #030A18; left aligned]: Tip: You can always ask Codex to do a retrospective and / update the Agents.md for you to avoid future mistakes or / get to an end result faster (Three lines, line pitch about 4.8% of slide height.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 57.2, y 15.6, w 42.8, h 68.8: 1024 by 463 pixels
- text at x 3.4, y 5.9, w 55.4, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: What AGENTS.md typically includes
- text at x 3.7, y 19.6, w 16.7, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: Common sections:
- text at x 4.5, y 26.4, w 1.2, h 3.5 [ArialMT; font size 3.46; #030A18]: ●
- text at x 7.5, y 26.4, w 41.1, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: Repo map: key directories, important modules
- text at x 4.5, y 31.1, w 1.2, h 3.5 [ArialMT; font size 3.46; #030A18]: ●
- text at x 7.5, y 31.1, w 39.7, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: How to run the project: install, build, test, lint
- text at x 4.5, y 35.9, w 1.2, h 3.5 [ArialMT; font size 3.46; #030A18]: ●
- text at x 7.5, y 35.9, w 48.4, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: Engineering guardrails: conventions, PR expectations,
- text at x 7.5, y 40.7, w 30.8, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: “do not” rules, security constraints
- text at x 4.5, y 45.4, w 1.2, h 3.5 [ArialMT; font size 3.46; #030A18]: ●
- text at x 7.5, y 45.4, w 47.2, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: How work gets veriﬁed: what “done” means and how
- text at x 7.5, y 50.2, w 11.1, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: to validate it
- text at x 3.7, y 63.7, w 50.1, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: Tip: You can always ask Codex to do a retrospective and
- text at x 3.7, y 68.5, w 51.2, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: update the Agents.md for you to avoid future mistakes or
- text at x 3.7, y 73.3, w 23.0, h 3.5 [OpenAISans-Medium; font size 3.46; #030A18]: get to an end result faster

### Slide 27

**Review notes**

- What would change the score: Merge or shrink the four repeated Q&A pages, and add a short stated takeaway to the dense text slides 23 and 27 so the audience sees the point rather than reading paragraphs.

**Text**

AGENTS.md best practices
Keep it brief and actionable
Commands, paths, rules, and pitfalls. Skip
generic explanations. Most of OpenAI’s
AGENTS.md files are less than 100 lines
Point to task-specific .md files
Reference supporting markdown (.md) files in
main Agents.md for specific workflows and tasks:
- Plans.md for planning general large tasks
- Code_review.md for code review instructions
- Architecture.md for changes that require
architecture modifications
OpenAI
Confidential and proprietary.

**Layout**

Two-column layout on a white slide with the title at top left, a left column starting at about x 6% holding one icon-headed block, and a right column starting at about x 52% holding a second icon-headed block with a bullet list. The title sits at y 5.9% with a left margin of 3.4%. The footer has the OpenAI wordmark at bottom left (x 2.7%, y 94%) followed by the confidentiality line at x 8.2%. No slide number is visible.

**Visuals**

Two line icons, both black single-weight outline strokes with rounded joins. (1) A magnifying glass at x 6.2%, y 37.3%, about 2.2% wide by 4% high, placed left of the heading 'Keep it brief and actionable'. (2) A open folder outline at x 52.3%, y 32%, about 2.2% wide by 4% high, placed left of the heading 'Point to task-specific .md files'. The OpenAI wordmark in bold black sans-serif at bottom left, about 3% wide by 1.4% high. No photographs, charts or tables.

**Design**

Background: Solid white #FFFFFF. Plain white background (#FFFFFF) with no gradient, pattern, shadow or boxes. All text is black (#000000). Typeface is OpenAI Sans, a geometric grotesque sans-serif: Medium for the title and body text and Bold for the two headings. Title size about 5.9% of slide height, headings and body about 3.2% of slide height, footer about 1.5%. The footer text is set in a Helvetica Neue style regular sans-serif. Line and shape styles: the two icons use black outline strokes; no rules or filled shapes. Corner radii, shadows and gradients: none.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: full-slide white background fill (covers the entire slide, no stroke, no shadow)
- title at x 3.4, y 5.9, w 41.8, h 5.9 [OpenAI Sans, Medium, normal case, no letter-spacing; font size 5.93% of slide height; #000000; left aligned]: AGENTS.md best practices
- icon at x 6.2, y 37.3, w 2.2, h 4 [#000000]: magnifying glass outline icon (black single-weight outline stroke with rounded ends, no fill)
- subtitle at x 10, y 37.8, w 23.3, h 3.2 [OpenAI Sans, Bold, normal case, no letter-spacing; font size 3.21% of slide height; #000000; left aligned]: Keep it brief and actionable
- body at x 6.3, y 44.5, w 35, h 11.7 [OpenAI Sans, Medium, normal case, no letter-spacing; font size 3.21% of slide height; #000000; left aligned]: Commands, paths, rules, and pitfalls. Skip / generic explanations. Most of OpenAI’s / AGENTS.md files are less than 100 lines (line spacing about 4.2% of slide height)
- icon at x 52.3, y 32, w 2.2, h 4 [#000000]: open folder outline icon (black single-weight outline stroke with rounded joins, no fill)
- subtitle at x 56.2, y 32.8, w 25.6, h 3.2 [OpenAI Sans, Bold, normal case, no letter-spacing; font size 3.21% of slide height; #000000; left aligned]: Point to task-specific .md files
- body at x 52.4, y 39.5, w 41, h 7.4 [OpenAI Sans, Medium, normal case, no letter-spacing; font size 3.21% of slide height; #000000; left aligned]: Reference supporting markdown (.md) files in / main Agents.md for specific workflows and tasks: ('Agents.md' is underlined in the same colour as the surrounding text)
- bullet list at x 53.7, y 50.4, w 39.7, h 16 [OpenAI Sans, Medium, normal case, no letter-spacing; font size 3.21% of slide height; #000000; left aligned]: - Plans.md for planning general large tasks / - Code_review.md for code review instructions / - Architecture.md for changes that require / architecture modifications (bullet style is a short hyphen dash, text indented to about x 56.2%, wrapped line 'architecture modifications' aligned with text, line spacing about 4.2% of slide height, items separated by a larger gap after the body paragraph)
- logo at x 2.7, y 94.1, w 3, h 1.4 [#000000]: OpenAI wordmark (bold black wordmark, no background)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [Helvetica Neue, Regular, normal case, no letter-spacing; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- text at x 3.4, y 5.9, w 41.8, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: AGENTS.md best practices
- text at x 56.2, y 32.8, w 25.6, h 3.2 [OpenAISans-Bold; font size 3.21; #000000]: Point to task-speciﬁc .md ﬁles
- text at x 10.0, y 37.8, w 23.3, h 3.2 [OpenAISans-Bold; font size 3.21; #000000]: Keep it brief and actionable
- text at x 52.4, y 39.5, w 38.3, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: Reference supporting markdown (.md) ﬁles in
- text at x 52.4, y 43.7, w 41.0, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: main Agents.md for speciﬁc workﬂows and tasks:
- text at x 6.3, y 44.5, w 35.0, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: Commands, paths, rules, and pitfalls. Skip
- text at x 6.3, y 48.7, w 32.8, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: generic explanations. Most of OpenAI’s
- text at x 53.7, y 50.4, w 36.5, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: - Plans.md for planning general large tasks
- text at x 6.3, y 53.0, w 32.9, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: AGENTS.md ﬁles are less than 100 lines
- text at x 53.7, y 54.7, w 39.7, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: - Code_review.md for code review instructions
- text at x 53.7, y 58.9, w 36.7, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: - Architecture.md for changes that require
- text at x 56.2, y 63.2, w 21.9, h 3.2 [OpenAISans-Medium; font size 3.21; #000000]: architecture modiﬁcations
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 28

**Text**

Q&A
OpenAI

**Layout**

Plain white slide with the title Q&A top-left and the OpenAI wordmark top-right on one header line. A full-width rounded image box sits below the header, spanning nearly the full slide width with equal side margins, and there is no footer, slide number or other text.

**Visuals**

One photograph-style abstract image in a rounded rectangle, occupying the lower part of the slide. Subject: a soft-focus, blurred abstract surface with light streaks, showing a diagonal transition from pale white-lavender at the upper left through light sky blue to saturated blue and violet at the right and bottom right. Composition: landscape crop, full image visible within a box with slightly rounded corners, no overlay. Lighting: soft, diffuse, luminous highlights at upper left and a bright sheen at top right. Dominant colours: white, periwinkle #A9B8F5, royal blue #4F6BEA, indigo #2F34D6. The OpenAI logo is a black wordmark, not an image element.

**Design**

Background: Solid white #FFFFFF. White background with black text. Title Q&A is a light-weight geometric sans-serif (Arial-like) in black #000000, about 8.9% of slide height for capitals. The OpenAI wordmark is a bold geometric sans-serif in black #000000 at top right. Image box has small corner radius (about 0.5% of slide width) and no shadow or border.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Background fill, white rectangle covering the slide (Full-bleed solid fill, no outline)
- photo at x 2.7, y 17, w 94.6, h 83 [#4F6BEA]: Blurred abstract gradient image with light streaks, diagonal transition from pale lavender-white at upper left to blue and indigo at right (Rounded rectangle crop with small corner radius, no border, no shadow, no overlay, full image visible, 1920 by 1080 pixels)
- title at x 3.6, y 5.8, w 10.5, h 8.9 [ArialMT, regular weight, sans-serif, capitals; font size 8.89% of slide height; #000000; left aligned]: Q&A
- logo at x 89, y 4.5, w 7.7, h 3.5 [bold geometric sans-serif; #000000; right aligned]: OpenAI wordmark (Black wordmark in top-right corner)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 2.7, y 17.0, w 94.6, h 83.0: 1920 by 1080 pixels
- text at x 3.6, y 5.8, w 10.5, h 8.9 [ArialMT; font size 8.89; #000000]: Q&A

### Slide 29

**Text**

OpenAI
How Teams are
Using Codex
Understand the art of the
possible

**Layout**

Two-column layout on a white slide. The OpenAI logo sits at top left, and the title and subtitle are left-aligned in the left column with a left margin of about 2.3% of the width. The title block starts at about 15.5% of the height. A single rounded-corner image fills the right column, from about 24.4% to the right edge with a right margin of about 1.6%, and runs from 4.8% to 95.5% of the height. No footer, page number or source line is visible.

**Visuals**

One photograph-style abstract image on the right. Subject: a soft, out-of-focus gradient with no people, objects or text. Composition: portrait-oriented rectangle, full frame, with small rounded corners, boxed rather than full-bleed. Lighting: diffuse and even. Dominant colours: pale lavender-white at the upper left (about #EEF0FD), periwinkle blue (about #6F8DF5) through the centre, royal blue (about #4A5EF0) at the lower left, and violet-indigo (about #3B3FD9) at the lower right. The image has no overlay or tint and is not cropped in any visible way. Also one OpenAI wordmark logo at top left, in black (#000000).

**Design**

Background: Solid white #FFFFFF. White (#FFFFFF) background with no gradient or pattern on the slide itself. Title is black (#000000) in a regular-weight Arial-style grotesque sans-serif, about 10.1% of the slide height per line, with tight line spacing. Subtitle is in OpenAI Sans SemiBold, a geometric grotesque sans-serif, about 4.2% of the slide height, in royal blue (#3442E0), set on two lines with generous leading. Logo is a black wordmark. The image has rounded corners of about 0.5% of the slide width, no shadow, and no border.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- logo at x 2.7, y 4.6, w 7.9, h 3.9 [#000000; left aligned]: OpenAI wordmark logo in black (Black wordmark on white, no container.)
- title at x 2.3, y 15.5, w 41, h 22.3 [ArialMT, regular weight, normal case, no letter-spacing; font size 10.12% of slide height; #000000; left aligned]: How Teams are / Using Codex (Two lines, line spacing about 12% of the slide height.)
- subtitle at x 2.8, y 41.2, w 28.3, h 10 [OpenAISans-SemiBold, normal case, no letter-spacing; font size 4.2% of slide height; #3442E0; left aligned]: Understand the art of the / possible (Two lines, line spacing about 5.7% of the slide height.)
- photo at x 24.4, y 4.8, w 75.6, h 90.7 [#4A5EF0]: Abstract blurred gradient image: pale lavender-white upper left fading into periwinkle, royal blue and violet-indigo toward the lower right, soft diffuse lighting, no people or objects (Boxed, not full-bleed; rounded corners of about 0.5% of slide width; no shadow, border or overlay; full frame, not cropped; 1920 by 1080 pixel source.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 24.4, y 4.8, w 75.6, h 90.7: 1920 by 1080 pixels
- text at x 2.3, y 15.5, w 41.0, h 10.1 [ArialMT; font size 10.12; #000000]: How Teams are
- text at x 2.3, y 27.7, w 32.6, h 10.1 [ArialMT; font size 10.12; #000000]: Using Codex
- text at x 2.8, y 41.2, w 28.3, h 4.2 [OpenAISans-SemiBold; font size 4.20; #3442E0]: Understand the art of the
- text at x 2.8, y 46.9, w 9.1, h 4.2 [OpenAISans-SemiBold; font size 4.20; #3442E0]: possible

### Slide 30

**Text**

Codex is deeply integrated with team workflows
Code Review
Auto and reactive code reviews to help
accelerate the review process and prevent
issues in production
Slack and Linear Mentions
Tag @Codex in a Slack channel or Linear
issue and it will gather context, choose the
right environment, and answer with a link
to the completed task
Codex SDK
Build your own integrations with the Codex
SDK. Auto-fix CI failures or kick off root-cause
analysis when an incident is declared
OpenAI
Confidential and proprietary.
30

**Layout**

Full-width title band across the top on a blue gradient, left-aligned at about 2.7% from the left edge with the title sitting near the top at 4.3% from the top. Three equal screenshot cards sit in a row beneath the title, each about 30 to 36% wide, with three matching text columns below them in white space starting at x 2.8%, 35.4% and 67.6%; body text runs in left-aligned lines of about 25 to 29% width. Footer sits at the bottom with the OpenAI wordmark at the left margin, a confidentiality line beside it, and the page number at the right margin at about 94.5% from the top.

**Visuals**

Three screenshot-style product images, left to right. (1) A code review view on a white card showing file path components/theme-provider.tsx, a green-highlighted diff with added lines 17 to 30 containing useEffect, localStorage, and classList.toggle code, followed by a bot comment chatgpt-codex-connector titled [P1] Prevent theme persistence effect from undoing saved dark mode with a paragraph of body text and a thumbs up reaction counter showing 1. (2) A Slack thread on a white card with Greg Brockman, Han Lu, Steve Lee and Codex messages, a code block reading still waiting for /root/code/openai/lib/strawberry, a 33 replies divider, a Codex summary with bullet points and a Testing line showing a green check and SKIP_TOR, and a View task link. (3) A dark terminal window with red, yellow and green traffic-light dots at top left, showing a YAML workflow named Codex Auto-Fix CI on Failure with keys on, workflow_run, jobs, auto-fix, and steps using openai/codex-action@main and a commit-message string in yellow. All three are uncropped rectangles with small corner radii and soft shadows, sitting on the gradient band. No logos besides the OpenAI wordmark in the footer.

**Design**

Background: Gradient from blue #3B3FE0 to pale lavender #C7D2FE across the top 88.9% of the slide, with plain white #FFFFFF below. Top band uses a blue to lavender gradient, with saturated blue #3B3FE0 at left shifting to lighter periwinkle #C7D2FE at top right. The lower area is plain white #FFFFFF. Title is white OpenAI Sans regular, about 5.4% of slide height. Column headings are black OpenAI Sans regular at about 3.5% of slide height, body text black OpenAI Sans regular at about 2.5%, with line spacing of about 3.3% between lines. Footer text is black HelveticaNeue at about 1.5%. Screenshot cards have rounded corners of roughly 1% of width and soft grey drop shadows; the terminal card is dark slate #263440 with white and coloured code text.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- photo at x 0, y 0, w 100, h 88.9 [#3B3FE0]: Background gradient image covering the top band of the slide, blue to lavender with no pictured content (Full-bleed raster, 2048 by 1344 pixels, sits behind all other elements; bottom edge meets the white area at 88.9% of height)
- shape at x 2.8, y 12.5, w 29.6, h 44.4 [#FFFFFF]: White rectangular card behind first screenshot (Rounded corners about 1% of width, faint shadow, sits beneath the screenshot)
- shape at x 35.5, y 12.2, w 29.4, h 45 [#FFFFFF]: White rectangular card behind second screenshot (Rounded corners about 1% of width, faint shadow, sits beneath the screenshot)
- shape at x 67.8, y 12.2, w 29.4, h 45 [#FFFFFF]: White rectangular card behind third screenshot (Rounded corners about 1% of width, faint shadow, sits beneath the screenshot)
- photo at x 2.9, y 14.4, w 29.4, h 42.4 [#FFFFFF]: Screenshot of a code review diff for components/theme-provider.tsx with a bot review comment (Raster 454 by 369 pixels, boxed inside its white card, not cropped; green added-line highlight #D4F5D4 with dark monospace code)
- photo at x 35.9, y 13.8, w 36.7, h 41.9 [#FFFFFF]: Screenshot of a Slack thread with Codex summary and View task link (Raster 493 by 316 pixels, boxed inside its white card, right edge cropped by the card edge; text rendered in black and grey)
- photo at x 67.8, y 12, w 32.2, h 45.5 [#263440]: Screenshot of a dark terminal window showing a YAML workflow (Raster 765 by 465 pixels, boxed with rounded corners, traffic-light dots red #FF5F57, yellow #FEBC2E, green #28C840 at top left; YAML keys in light blue and purple, strings in yellow)
- text at x 2.7, y 4.3, w 66.6, h 5.4 [OpenAI Sans, regular, sans-serif; font size 5.43% of slide height; #FFFFFF; left aligned]: Codex is deeply integrated with team workflows (Single line, normal case, no letter-spacing change)
- text at x 6, y 44.1, w 13.9, h 2.7 [HelveticaNeue, regular, sans-serif; font size 2.72% of slide height; #000000; left aligned]: Weekly Active Users (Present in the file but not visible in the rendered slide; it lies under the screenshot cards, so it is hidden)
- text at x 2.8, y 66.9, w 11.6, h 3.5 [OpenAI Sans, regular, sans-serif; font size 3.46% of slide height; #000000; left aligned]: Code Review (Column heading)
- text at x 2.8, y 72.8, w 27, h 8.9 [OpenAI Sans, regular, sans-serif; font size 2.47% of slide height; #000000; left aligned]: Auto and reactive code reviews to help / accelerate the review process and prevent / issues in production (Body paragraph, three lines, line pitch about 3.3% of height)
- text at x 35.4, y 66.9, w 23.5, h 3.5 [OpenAI Sans, regular, sans-serif; font size 3.46% of slide height; #000000; left aligned]: Slack and Linear Mentions (Column heading)
- text at x 35.4, y 72.8, w 27.1, h 12.3 [OpenAI Sans, regular, sans-serif; font size 2.47% of slide height; #000000; left aligned]: Tag @Codex in a Slack channel or Linear / issue and it will gather context, choose the / right environment, and answer with a link / to the completed task (Body paragraph, four lines, line pitch about 3.3% of height)
- text at x 67.6, y 66.9, w 10, h 3.5 [OpenAI Sans, regular, sans-serif; font size 3.46% of slide height; #000000; left aligned]: Codex SDK (Column heading)
- text at x 67.6, y 72.8, w 29.1, h 8.9 [OpenAI Sans, regular, sans-serif; font size 2.47% of slide height; #000000; left aligned]: Build your own integrations with the Codex / SDK. Auto-fix CI failures or kick off root-cause / analysis when an incident is declared (Body paragraph, three lines, line pitch about 3.3% of height)
- logo at x 2.7, y 94.2, w 6, h 1.5 [#000000]: OpenAI wordmark (Bold black wordmark at the bottom left, position estimated from the render)
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue, regular, sans-serif; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.
- page number at x 96.4, y 94.4, w 0.9, h 1.5 [HelveticaNeue, regular, sans-serif; font size 1.48% of slide height; #000000; right aligned]: 30

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 67.8, y 12.2, w 29.4, h 45.0: #FFFFFF
- fill at x 35.5, y 12.2, w 29.4, h 45.0: #FFFFFF
- fill at x 2.8, y 12.5, w 29.6, h 44.4: #FFFFFF
- image at x 0.0, y 0.0, w 100.0, h 88.9: 2048 by 1344 pixels (covers the slide)
- image at x 67.8, y 12.0, w 32.2, h 45.5: 765 by 465 pixels
- image at x 35.9, y 13.8, w 36.7, h 41.9: 493 by 316 pixels
- image at x 2.9, y 14.4, w 29.4, h 42.4: 454 by 369 pixels
- text at x 2.7, y 4.3, w 66.6, h 5.4 [OpenAISans-Regular; font size 5.43; #FFFFFF]: Codex is deeply integrated with team workﬂows
- text at x 6.0, y 44.1, w 13.9, h 2.7 [HelveticaNeue; font size 2.72; #000000]: Weekly Active Users
- text at x 2.8, y 66.9, w 11.6, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Code Review
- text at x 35.4, y 66.9, w 23.5, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Slack and Linear Mentions
- text at x 67.6, y 66.9, w 10.0, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Codex SDK
- text at x 2.8, y 72.8, w 24.8, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Auto and reactive code reviews to help
- text at x 35.4, y 72.8, w 25.9, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Tag @Codex in a Slack channel or Linear
- text at x 67.6, y 72.8, w 27.4, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Build your own integrations with the Codex
- text at x 2.8, y 76.1, w 27.0, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: accelerate the review process and prevent
- text at x 35.4, y 76.1, w 27.1, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: issue and it will gather context, choose the
- text at x 67.6, y 76.1, w 29.1, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: SDK. Auto-ﬁx CI failures or kick off root-cause
- text at x 2.8, y 79.3, w 12.6, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: issues in production
- text at x 35.4, y 79.3, w 26.3, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: right environment, and answer with a link
- text at x 67.6, y 79.3, w 23.4, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: analysis when an incident is declared
- text at x 35.4, y 82.6, w 13.9, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: to the completed task
- text at x 96.4, y 94.4, w 0.9, h 1.5 [HelveticaNeue; font size 1.48; #000000]: 30
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 31

**Text**

What we’ve seen with OpenAI’s Engineers
●
Some teams are entirely “hands off” - they
prompt everything and have Codex write all
code
●
Codex reviews 100% of PRs internally
●
Agents work proactively in the background
●
Early versions of 5.3-Codex debugged its own
training runs
OpenAI
Confidential and proprietary.
31

**Layout**

Two-column layout on a white field with a left title and a left-aligned bulleted list in the left column, and an image in the right column. The title sits top-left at about x 3.7%, y 8.2%, and the bullets start at x 4.3% (bullet glyphs) and x 7.5% (text), vertically spread from y 30.4% to 71.6%. The image occupies x 56.1% to 89.9% and y 22.2% to 77.9%. The footer has the OpenAI wordmark bottom-left at about x 2.7%, y 94.5%, a 'Confidential and proprietary.' line at x 8.2%, and the page number 31 at bottom-right, x 96.5%, y 94.4%.

**Visuals**

One image on the right, placed as a single raster image (388 by 359 pixels, not a photograph), boxed with square corners and no border. Its top part shows the black sans-serif heading 'How OpenAI' / 'uses Codex' on white, and below it a blurred pastel gradient panel. The gradient runs from sky blue at the left and bottom through lavender and violet in the centre, to yellow-green, orange and pink at upper right, with faint monospace code fragments in pale green and white scattered across it (for example fragments such as 'def softmax', 'reduce', 'axis=', 'dynamic', 'return'). Centred over the lower part of the gradient is a white rounded card with a soft shadow containing a light-lavender pill with a purple git-merge branch icon and the purple text 'Merged', followed by green '+217' and red '-196' values. No chart, table or logo other than the footer wordmark.

**Design**

Background: Solid white #FFFFFF. Plain white background with no gradient or pattern. Title in black OpenAI Sans Medium at about 5.9% of slide height, body in black OpenAI Sans Regular at about 3.95% of slide height, footer and page number at about 1.5%. Body text is black #000000; the only accent colours come from the image (sky blue, lavender, green, orange, pink, purple #8B5CF6 for the Merged pill, green #2E9E4F and red #D93A3A for the diff counts). Bullets are solid black round glyphs in Arial, with generous line spacing (about 1.5 lines) and about 6% gaps between bullet items. No rules, shapes or shadows apart from the soft shadow on the card inside the image.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide white background fill (Covers the entire slide, no border.)
- photo at x 56.1, y 22.2, w 33.8, h 55.7: Illustration image 'How OpenAI uses Codex' with blurred gradient panel, code fragments and a Merged diff card (+217 -196) (Placed raster image 388 by 359 px, square corners, no border, not full-bleed, not cropped. Contains heading text in black sans-serif above a pastel gradient panel (sky blue, lavender, green, orange, pink) with pale code fragments and a white rounded card with soft shadow, lavender Merged pill with purple branch icon, green +217 and red -196.)
- title at x 3.7, y 8.2, w 63.8, h 5.9 [OpenAI Sans, Medium weight, sans-serif; font size 5.93% of slide height; #000000; left aligned]: What we’ve seen with OpenAI’s Engineers
- body at x 4.3, y 30.4, w 1.3, h 4 [Arial, Regular, sans-serif; font size 3.95% of slide height; #000000; left aligned]: ● (Solid round bullet glyph, first item.)
- bullet list at x 7.5, y 30.4, w 44.7, h 14.5 [OpenAI Sans, Regular weight, sans-serif; font size 3.95% of slide height; #000000; left aligned]: Some teams are entirely “hands off” - they / prompt everything and have Codex write all / code (Line spacing about 5.2% of slide height between baselines. Bullet item one, three lines.)
- body at x 4.3, y 48.1, w 1.3, h 4 [Arial, Regular, sans-serif; font size 3.95% of slide height; #000000; left aligned]: ● (Solid round bullet glyph, second item.)
- bullet list at x 7.5, y 48.1, w 37.9, h 4 [OpenAI Sans, Regular weight, sans-serif; font size 3.95% of slide height; #000000; left aligned]: Codex reviews 100% of PRs internally
- body at x 4.3, y 55.2, w 1.3, h 4 [Arial, Regular, sans-serif; font size 3.95% of slide height; #000000; left aligned]: ● (Solid round bullet glyph, third item.)
- bullet list at x 7.5, y 55.2, w 43.3, h 4 [OpenAI Sans, Regular weight, sans-serif; font size 3.95% of slide height; #000000; left aligned]: Agents work proactively in the background
- body at x 4.3, y 62.4, w 1.3, h 4 [Arial, Regular, sans-serif; font size 3.95% of slide height; #000000; left aligned]: ● (Solid round bullet glyph, fourth item.)
- bullet list at x 7.5, y 62.4, w 46.8, h 9.2 [OpenAI Sans, Regular weight, sans-serif; font size 3.95% of slide height; #000000; left aligned]: Early versions of 5.3-Codex debugged its own / training runs (Bullet item four, two lines.)
- page number at x 72.2, y 70.6, w 0.8, h 1.5 [OpenAI Sans, Regular weight, sans-serif; font size 1.48% of slide height; #000000; left aligned]: 31 (Text box recorded in the file at this position, behind the image; not visibly rendered on the slide.)
- logo at x 2.7, y 94.5, w 3, h 1.5 [#000000]: OpenAI wordmark (Small bold black sans-serif OpenAI wordmark at bottom-left, footer.)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans, Regular weight, sans-serif; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.
- page number at x 96.5, y 94.4, w 0.8, h 1.5 [OpenAI Sans, Regular weight, sans-serif; font size 1.48% of slide height; #000000; right aligned]: 31

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 56.1, y 22.2, w 33.8, h 55.7: 388 by 359 pixels
- text at x 3.7, y 8.2, w 63.8, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: What we’ve seen with OpenAI’s Engineers
- text at x 4.3, y 30.4, w 1.3, h 4.0 [ArialMT; font size 3.95; #000000]: ●
- text at x 7.5, y 30.4, w 43.4, h 4.0 [OpenAISans-Regular; font size 3.95; #000000]: Some teams are entirely “hands off ” - they
- text at x 7.5, y 35.6, w 44.7, h 4.0 [OpenAISans-Regular; font size 3.95; #000000]: prompt everything and have Codex write all
- text at x 7.5, y 40.9, w 5.0, h 4.0 [OpenAISans-Regular; font size 3.95; #000000]: code
- text at x 4.3, y 48.1, w 1.3, h 4.0 [ArialMT; font size 3.95; #000000]: ●
- text at x 7.5, y 48.1, w 37.9, h 4.0 [OpenAISans-Regular; font size 3.95; #000000]: Codex reviews 100% of PRs internally
- text at x 4.3, y 55.2, w 1.3, h 4.0 [ArialMT; font size 3.95; #000000]: ●
- text at x 7.5, y 55.2, w 43.3, h 4.0 [OpenAISans-Regular; font size 3.95; #000000]: Agents work proactively in the background
- text at x 4.3, y 62.4, w 1.3, h 4.0 [ArialMT; font size 3.95; #000000]: ●
- text at x 7.5, y 62.4, w 46.8, h 4.0 [OpenAISans-Regular; font size 3.95; #000000]: Early versions of 5.3-Codex debugged its own
- text at x 7.5, y 67.6, w 12.8, h 4.0 [OpenAISans-Regular; font size 3.95; #000000]: training runs
- text at x 72.2, y 70.6, w 0.8, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: 31
- text at x 96.5, y 94.4, w 0.8, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: 31
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 32

**Review notes**

- Worth studying: Large headline metrics with a single customer quote give the proof points visual weight and a clear hierarchy.

**Text**

Cisco is transforming its
engineering org with Codex
“
The biggest gains came
when we stopped thinking
about Codex as a tool and
started treating it as part
of the team.”
Ryan Brady
Principal Engineer
1,500
Engineering hours
saved per month
10–15x
Faster defect remediation
20%
Reduction in
build times
OpenAI
Conﬁdential and proprietary.
32

**Layout**

Three-column grid on a white slide with about 2.7% side margins. The title sits top-left in two lines, with the Cisco logo top-right. Three content panels sit in the middle band: a tall quote card on the left, two stacked stat cards in the centre column, and a tall gradient card on the right, all aligned to a common top edge at about 29.8% and bottom edge at about 90.9%. The footer row at about 94.4% holds the OpenAI wordmark and confidentiality line on the left and the slide number at the far right.

**Visuals**

Four rectangular panels and one logo. (1) Cisco logo, top-right: the black Cisco wordmark with its bar-pattern symbol above the letters CISCO, set on white. (2) Left quote card: light blue rounded rectangle holding a large blue opening quotation mark, a five-line quote from Ryan Brady, and his name and title. (3) Centre top card: light blue rounded rectangle with the stat 1,500 in blue and the label Engineering hours saved per month. (4) Centre bottom card: light blue rounded rectangle with the stat 10–15x in blue and the label Faster defect remediation. (5) Right tall card: a full-card photograph-style gradient fill with soft blurred blue and violet tones, lighter lavender at the top-left corner and deeper indigo-blue at the bottom-right. It has no visible subject, is not cropped to a photo, and carries the white stat 20% and label Reduction in build times over its lower part. No charts or tables.

**Design**

Background: White #FFFFFF.

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 35.1, y 61.3, w 29.8, h 29.1: #A3BEFA, opacity 0.25
- fill at x 35.1, y 29.8, w 29.8, h 29.1: #A3BEFA, opacity 0.25
- fill at x 2.7, y 29.8, w 29.8, h 60.6: #A3BEFA, opacity 0.25
- image at x 18.2, y 22.3, w 81.8, h 77.7: 1920 by 1080 pixels (covers the slide)
- image at x 89.5, y 4.8, w 7.8, h 7.3: 128 by 68 pixels
- text at x 2.7, y 4.3, w 34.7, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: Cisco is transforming its
- text at x 2.7, y 10.9, w 38.5, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: engineering org with Codex
- text at x 4.9, y 36.1, w 4.0, h 14.8 [OpenAISans-Bold; font size 14.81; #4285F4]: “
- text at x 37.5, y 35.9, w 12.5, h 8.9 [OpenAISans-Medium; font size 8.89; #4285F4]: 1,500
- text at x 70.6, y 60.3, w 13.4, h 11.1 [OpenAISans-Medium; font size 11.11; #FFFFFF]: 20%
- text at x 37.6, y 69.5, w 15.1, h 8.9 [OpenAISans-Medium; font size 8.89; #4285F4]: 10–15x
- text at x 37.5, y 45.8, w 16.6, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Engineering hours
- text at x 70.6, y 72.8, w 18.4, h 5.4 [OpenAISans-Medium; font size 5.43; #FFFFFF]: Reduction in
- text at x 37.5, y 49.9, w 14.9, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: saved per month
- text at x 70.6, y 79.2, w 15.5, h 5.4 [OpenAISans-Medium; font size 5.43; #FFFFFF]: build times
- text at x 5.1, y 51.2, w 21.6, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: The biggest gains came
- text at x 5.1, y 55.8, w 23.7, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: when we stopped thinking
- text at x 5.1, y 60.4, w 23.5, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: about Codex as a tool and
- text at x 5.1, y 65.1, w 22.7, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: started treating it as part
- text at x 5.1, y 69.7, w 11.3, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: of the team.”
- text at x 37.6, y 79.4, w 22.9, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Faster defect remediation
- text at x 5.1, y 78.8, w 7.6, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Ryan Brady
- text at x 5.1, y 82.1, w 11.6, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Principal Engineer
- text at x 96.4, y 94.4, w 0.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: 32
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 33

**Text**

Datadog ships more reliably
with Codex Code Review
DATADOG
“
Codex changed my mind for what code review should
be. It’s not about replicating our best human
reviewers. It's about finding critical flaws and edge
cases that humans struggle to see when reviewing
changes in isolation.”
Brad Carter
Engineering Manager at Datadog
22%
of past incidents had risks
Codex flagged
1,000+
DataDog engineers
using Codex weekly
Confidential and proprietary.
33

**Layout**

Two-column card grid on a white field. The title sits top-left in two lines, with the DATADOG logo top-right, both aligned to a left margin of about 2.7% and a right margin of about 2.7%. A large left card spans the full content height, and two stacked cards sit on the right, all with equal gutters. A small footer row at the bottom holds the OpenAI wordmark and confidentiality line on the left and the page number at the far right.

**Visuals**

Four non-text visual items. (1) Blue gradient image (1920 by 1080 px) occupying the right portion of the slide from about x 27.5% to the right edge and y 0% to about 94%; its blue gradient, from deep indigo #4F5BE0 at lower left through royal blue #4285F4 to pale periwinkle #C9D6FB at upper right, is visible behind the '22%' card. (2) Blue quotation mark glyph in #4285F4, a heavy sans-serif double quote, at top-left of the left card. (3) DATADOG logo in black: a stylised dog head inside a rounded black square outline on the left, with the bold wordmark 'DATADOG' to its right. (4) OpenAI wordmark in black, small bold sans-serif, bottom-left in the footer. No charts, tables or photographs.

**Design**

Background: Solid white #FFFFFF, with a blue gradient image shown in the right portion of the slide. Clean white slide with a minimal corporate look. Titles and body text are black #000000. Accents are Google blue #4285F4 for the quote mark and the '1,000+' figure. Card fills are pale blue #E8EFFD (#A3BEFA at 25% opacity over white). Typeface is a geometric grotesque sans-serif (OpenAI Sans), Medium weight for titles and large figures, Regular for body and captions. Title text is about 5.4% of slide height; body text about 3.5%; attribution about 2.5%; footer about 1.5%. Cards have small rounded corners of about 1% of slide width, no borders and no shadows. The blue gradient card has a soft diagonal gradient with no border.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- image at x 27.5, y 0, w 72.5, h 93.9: Blue gradient image (1920 by 1080 px), visible behind the 22% card (Full-frame gradient, not a photograph; soft diagonal blend from indigo #4F5BE0 to periwinkle #C9D6FB; the top-right card area is brightest.)
- shape at x 2.7, y 29.8, w 54.1, h 60.6 [#E8EFFD]: Left quote card, pale blue rounded rectangle (Fill #A3BEFA at 25% opacity over white; corner radius about 1% of slide width; no border or shadow.)
- shape at x 59.4, y 29.8, w 37.9, h 29.1 [#4285F4]: Right top stat card, blue gradient rounded rectangle (Filled with the blue gradient image; corner radius about 1% of slide width; no border or shadow.)
- shape at x 59.4, y 61.4, w 37.9, h 29.1 [#E8EFFD]: Right bottom stat card, pale blue rounded rectangle (Fill #A3BEFA at 25% opacity over white; corner radius about 1% of slide width; no border or shadow.)
- title at x 2.7, y 4.3, w 39.8, h 12 [OpenAI Sans, Medium, sans-serif; font size 5.43% of slide height; #000000; left aligned]: Datadog ships more reliably / with Codex Code Review (Two lines, line spacing about 6.6% of slide height.)
- logo at x 84.5, y 5, w 14.8, h 6.5 [#000000]: DATADOG logo: stylised dog head in a black rounded-square outline, with the 'DATADOG' wordmark in bold black capitals (Dog icon about 4.5% wide at left of the box; wordmark about 10.5% wide at right.)
- icon at x 4.9, y 36.1, w 4, h 14.8 [OpenAI Sans, Bold, sans-serif; font size 14.81% of slide height; #4285F4; left aligned]: Blue opening quotation mark glyph (Heavy double-comma style quote mark, rendered as the glyph U+201C.)
- body at x 5.1, y 48.9, w 45.1, h 18.5 [OpenAI Sans, Regular, sans-serif; font size 3.46% of slide height; #000000; left aligned]: Codex changed my mind for what code review should / be. It’s not about replicating our best human / reviewers. It's about finding critical flaws and edge / cases that humans struggle to see when reviewing / changes in isolation.” (Pull quote, line spacing about 4.7% of slide height, ragged right.)
- body at x 5.1, y 78.8, w 21.1, h 5.8 [OpenAI Sans, Regular, sans-serif; font size 2.47% of slide height; #000000; left aligned]: Brad Carter / Engineering Manager at Datadog (Attribution, two lines with line spacing about 3.3% of slide height.)
- body at x 61.8, y 36.7, w 10.3, h 8.9 [OpenAI Sans, Medium, sans-serif; font size 8.89% of slide height; #FFFFFF; left aligned]: 22%
- body at x 61.8, y 46.9, w 23.6, h 7.6 [OpenAI Sans, Regular, sans-serif; font size 3.46% of slide height; #FFFFFF; left aligned]: of past incidents had risks / Codex flagged (Two lines, ragged right.)
- body at x 61.9, y 67.9, w 15.7, h 8.9 [OpenAI Sans, Medium, sans-serif; font size 8.89% of slide height; #4285F4; left aligned]: 1,000+
- body at x 61.9, y 77.8, w 17.5, h 7.5 [OpenAI Sans, Regular, sans-serif; font size 3.46% of slide height; #000000; left aligned]: DataDog engineers / using Codex weekly (Two lines, ragged right.)
- logo at x 2.7, y 94.3, w 5.2, h 1.5 [#000000]: OpenAI wordmark (Small bold black sans-serif wordmark in footer.)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans, Regular, sans-serif; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.
- page number at x 96.3, y 94.4, w 1, h 1.5 [OpenAI Sans, Regular, sans-serif; font size 1.48% of slide height; #000000; right aligned]: 33

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 59.4, y 61.4, w 37.9, h 29.1: #A3BEFA, opacity 0.25
- fill at x 2.7, y 29.8, w 54.1, h 60.6: #A3BEFA, opacity 0.25
- image at x 27.5, y 0.0, w 72.5, h 93.9: 1920 by 1080 pixels (covers the slide)
- text at x 2.7, y 4.3, w 39.8, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: Datadog ships more reliably
- text at x 2.7, y 10.9, w 34.7, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: with Codex Code Review
- text at x 4.9, y 36.1, w 4.0, h 14.8 [OpenAISans-Bold; font size 14.81; #4285F4]: “
- text at x 61.8, y 36.7, w 10.3, h 8.9 [OpenAISans-Medium; font size 8.89; #FFFFFF]: 22%
- text at x 61.9, y 67.9, w 15.7, h 8.9 [OpenAISans-Medium; font size 8.89; #4285F4]: 1,000+
- text at x 61.8, y 46.9, w 23.6, h 3.5 [OpenAISans-Regular; font size 3.46; #FFFFFF]: of past incidents had risks
- text at x 5.1, y 48.9, w 47.7, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Codex changed my mind for what code review should
- text at x 61.8, y 51.0, w 12.9, h 3.5 [OpenAISans-Regular; font size 3.46; #FFFFFF]: Codex ﬂagged
- text at x 5.1, y 53.5, w 39.6, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: be. It’s not about replicating our best human
- text at x 5.1, y 58.2, w 45.1, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: reviewers. It's about ﬁnding critical ﬂaws and edge
- text at x 5.1, y 62.8, w 45.0, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: cases that humans struggle to see when reviewing
- text at x 5.1, y 67.4, w 18.7, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: changes in isolation.”
- text at x 61.9, y 77.8, w 17.5, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: DataDog engineers
- text at x 61.9, y 81.8, w 17.4, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: using Codex weekly
- text at x 5.1, y 78.8, w 7.4, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Brad Carter
- text at x 5.1, y 82.1, w 21.1, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Engineering Manager at Datadog
- text at x 96.3, y 94.4, w 1.0, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: 33
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 34

**Text**

Rakuten fixes issues twice as fast
with Codex
Rakuten
“
We don’t just care about generating code quickly. We
care about shipping safely. Speed without safety is
not success. Our role is not to check every line of
code anymore. Our role is to define clearly what we
want and establish how to verify it.
Yusuke Kaji
General Manager of AI at Rakuten
50%
Reduction in Mean Time to
Repair (MTTR)
OpenAI
Confidential and proprietary.

**Layout**

Two-column layout on a white slide with 2.7% side margins. The title sits top-left across two lines, with the Rakuten logo top-right on the same band. A tall pale-blue quote panel fills the left column (about 54% wide), while the right column stacks a photograph above a blue statistic box, both aligned to the same top and bottom edges as the panel. The footer sits at the bottom left, with the OpenAI wordmark followed by the confidentiality line. No slide number is visible.

**Visuals**

One photograph of a man in a grey crew-neck sweater and glasses, seated with hands clasped and smiling slightly, in a warm dimly lit room with wooden shelves, potted plants, a red bowl, stacked books and a small lamp behind him. It is boxed with slightly rounded corners, warm brown and charcoal tones, soft ambient lighting, and framed at medium shot with the subject centred. A blue quote glyph (“) sits above the quote as an icon, in #4285F4. A Rakuten wordmark logo in dark charcoal #231F20 appears top-right. A blue statistic panel with a diagonal gradient from deep blue #4B67F0 at the lower left to pale periwinkle #B8CCFB at the upper right holds the large white figure 50%. A placed 1920 by 1080 image covers the full slide area at the position listed, but it shows no visible content on the rendered slide. The OpenAI wordmark appears bottom-left in black.

**Design**

Background: Solid white #FFFFFF. Plain white background. Title and body text are black #000000. Accent colour is blue #4285F4 for the quote glyph. The quote panel is pale blue #E9F0FD, which is #A3BEFA at 25% opacity over white. The statistic panel uses a blue gradient with white text. Typeface is OpenAI Sans, a geometric grotesque sans-serif similar to Inter. Title is Medium weight at about 5.4% of slide height, quote body is Regular at about 3.5%, name and role are Regular at about 2.5%, footer is Regular at about 1.5%, and the statistic figure is Medium at about 8.9%. Rounded corners of about 1.5% of slide width on the panels, no shadows, no visible borders.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide background fill (Covers the slide, no stroke)
- shape at x 2.7, y 18.9, w 54.1, h 71.5 [#E9F0FD]: Quote panel, rounded rectangle (Fill #A3BEFA at opacity 0.25 over white, corner radius about 1.5% of slide width, no stroke, no shadow)
- image at x 27.5, y 23.9, w 72.5, h 76.1: Placed 1920 by 1080 image, no visible content on rendered slide (Covers the slide area under the other elements; nothing visible)
- photo at x 59.4, y 18.9, w 37.2, h 40.1: Man in grey crew-neck sweater and glasses seated with hands clasped in a warm room with shelves, plants, bowl, books and lamp (Boxed, slightly rounded corners, warm brown and charcoal tones, medium shot, subject centred, not cropped)
- shape at x 59.4, y 61.4, w 37.9, h 29 [#4B67F0]: Statistic panel, rounded rectangle with blue gradient (Diagonal gradient from #4B67F0 at lower left to #B8CCFB at upper right, corner radius about 1% of slide width, no stroke, no shadow)
- logo at x 85.3, y 4.8, w 12, h 6.1 [#231F20]: Rakuten wordmark (Dark charcoal bold wordmark, top-right corner)
- title at x 2.7, y 4.3, w 47.4, h 12 [OpenAI Sans Medium; font size 5.43% of slide height; #000000; left aligned]: Rakuten fixes issues twice as fast / with Codex (Two lines, line spacing about 6.5% of slide height)
- icon at x 4.9, y 29.3, w 4, h 14.8 [#4285F4]: Blue opening quotation mark glyph (“) (Bold sans-serif quote glyph, solid fill, no outline)
- body at x 5.1, y 48.8, w 45.5, h 18.7 [OpenAI Sans Regular; font size 3.46% of slide height; #000000; left aligned]: We don’t just care about generating code quickly. We / care about shipping safely. Speed without safety is / not success. Our role is not to check every line of / code anymore. Our role is to define clearly what we / want and establish how to verify it. (Five lines, line spacing about 4.7% of slide height)
- body at x 5.1, y 78.7, w 21.4, h 5.8 [OpenAI Sans Regular; font size 2.47% of slide height; #000000; left aligned]: Yusuke Kaji / General Manager of AI at Rakuten (Two lines, name followed by role, line spacing about 3.3% of slide height)
- photo at x 59.4, y 61.4, w 37.9, h 29: Image box boundary of statistic panel (Structural duplicate boundary, no separate visible content)
- body at x 61.8, y 68.3, w 10.8, h 8.9 [OpenAI Sans Medium; font size 8.89% of slide height; #FFFFFF; left aligned]: 50%
- body at x 61.8, y 78.5, w 24.3, h 7.6 [OpenAI Sans Regular; font size 3.46% of slide height; #FFFFFF; left aligned]: Reduction in Mean Time to / Repair (MTTR) (Two lines, line spacing about 4.1% of slide height)
- logo at x 2.7, y 94.2, w 5.6, h 1.3 [#000000]: OpenAI wordmark (Small black wordmark, bottom-left)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans Regular; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary.

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- fill at x 2.7, y 18.9, w 54.1, h 71.5: #A3BEFA, opacity 0.25
- image at x 27.5, y 23.9, w 72.5, h 76.1: 1920 by 1080 pixels (covers the slide)
- image at x 59.4, y 18.9, w 37.2, h 40.1: 485 by 294 pixels
- image at x 85.3, y 4.8, w 12.0, h 6.1: 206 by 61 pixels
- text at x 2.7, y 4.3, w 47.4, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: Rakuten ﬁxes issues twice as fast
- text at x 2.7, y 10.9, w 15.8, h 5.4 [OpenAISans-Medium; font size 5.43; #000000]: with Codex
- text at x 4.9, y 29.3, w 4.0, h 14.8 [OpenAISans-Bold; font size 14.81; #4285F4]: “
- text at x 61.8, y 68.3, w 10.8, h 8.9 [OpenAISans-Medium; font size 8.89; #FFFFFF]: 50%
- text at x 5.1, y 48.8, w 47.3, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: We don’t just care about generating code quickly. We
- text at x 5.1, y 53.5, w 45.2, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: care about shipping safely. Speed without safety is
- text at x 5.1, y 58.1, w 43.4, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: not success. Our role is not to check every line of
- text at x 5.1, y 62.7, w 45.4, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: code anymore. Our role is to deﬁne clearly what we
- text at x 5.1, y 67.4, w 30.7, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: want and establish how to verify it.
- text at x 61.8, y 78.5, w 24.3, h 3.5 [OpenAISans-Regular; font size 3.46; #FFFFFF]: Reduction in Mean Time to
- text at x 61.8, y 82.6, w 13.6, h 3.5 [OpenAISans-Regular; font size 3.46; #FFFFFF]: Repair (MTTR)
- text at x 5.1, y 78.7, w 7.1, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: Yusuke Kaji
- text at x 5.1, y 82.0, w 21.4, h 2.5 [OpenAISans-Regular; font size 2.47; #000000]: General Manager of AI at Rakuten
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 35

**Text**

Shipping Sora in / 28 days with Codex / Lean team of 4 engineers / 18-day to prototype and 10 days to launch / Codex wrote ~85% of the code, humans / focused on architecture and review / Parallel Codex workstreams: debug, / refactor, tests, reliability / 99.9% crash-free, 4.9/5, #1 on Google Play / Confidential and proprietary. / 35 / OpenAI

**Layout**

Two-column layout on a white field. The square-cornered image panel sits in the left column, from about 2.7% to 49.4% of the slide width and 6.7% to 89.7% of its height. The title and bullet list sit in the right column, starting at about 53.8% of the width, with the title top at about 13.3% of the height, left-aligned. The footer wordmark and confidentiality line sit along the bottom at about 94.4% of the height, with the page number at the bottom right.

**Visuals**

One illustration on the left: a rounded-corner square image (1024 by 1024 pixels, shown at about 46.7% of slide width) of a night-sky scene. Deep navy blue (#14206B) at the top fades into a bright iridescent band of pink, violet, cyan and pale yellow sweeping diagonally from lower left to upper right, with white and pale-blue sparkles and clouds at the bottom. Faint monospace code fragments are scattered in the sky. Overlaid in the centre-left is a frosted-glass UI card, about 34% of slide width, with rounded corners and a white translucent fill tinted with pink and blue from the background. Its top row shows a dark navy git-branch icon, the text 'sora/android' in a dark charcoal sans-serif at large size, and a blue (#4FC3F7) downward chevron at the right. Below is a white inner panel with a lavender (#E4D9FA) pill containing a purple (#6B3FD1) git-merge icon and the word 'Merged'. To its right are '+217' in green (#3E8E4E) and '-196' in red (#C84B55). The photo is not cropped, full-bleed, or tinted apart from the glass overlay. No chart, table or logo appears in the image.

**Design**

Background: Solid white #FFFFFF. Plain white background (#FFFFFF). Titles and body text are black (#000000). Accent colours appear only in the image: navy #14206B, lavender #E4D9FA, purple #6B3FD1, green #3E8E4E and red #C84B55. Title is OpenAI Sans Medium, a geometric grotesque sans-serif, at about 5.9% of slide height, with normal tracking. Body is OpenAI Sans Regular at about 3.5% of slide height, with bold runs for the figures. Bullets are solid black dots in Arial. Image corners are rounded at about 0.6% of slide width. No shadows, gradients, rules or other shapes are present on the slide.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- photo at x 2.7, y 6.7, w 46.7, h 83 [#14206B]: Night-sky illustration with iridescent rainbow band, rounded-corner square image 1024 by 1024 pixels (Dominant fill navy #14206B to pink, violet and cyan. Rounded corners about 0.6% of slide width. Not cropped, not boxed with a border, no tint apart from the image itself. Overlaps the glass card.)
- shape at x 8.8, y 35.4, w 34.4, h 27.6 [#F4F1FA]: Frosted-glass UI card showing sora/android, git-branch icon, chevron, and a Merged pill with +217 and -196 (Translucent white rounded rectangle, corner radius about 2% of slide width, soft white edge and slight pink and blue tint from background. Sits on top of the photo. Contains the text sora/android in dark charcoal sans-serif, a blue chevron #4FC3F7, and an inner white panel at x 10.5, y 49.3, w 31.0, h 10.8 with a lavender pill #E4D9FA holding a purple #6B3FD1 Merged label, then +217 in green #3E8E4E and -196 in red #C84B55.)
- title at x 53.8, y 13.3, w 29.8, h 12.8 [OpenAI Sans, Medium weight, sans-serif, normal case, normal letter-spacing; font size 5.93% of slide height; #000000; left aligned]: Shipping Sora in / 28 days with Codex (Two lines, line spacing about 1.2 times the font size. First line 'Shipping Sora in', second line '28 days with Codex'.)
- bullet list at x 54.9, y 32.2, w 45, h 47 [OpenAI Sans, Regular weight, sans-serif, normal case, normal letter-spacing; bold runs for ~85% and 99.9%; font size 3.46% of slide height; #000000; left aligned]: ● Lean team of 4 engineers / ● 18-day to prototype and 10 days to launch / ● Codex wrote ~85% of the code, humans / focused on architecture and review / ● Parallel Codex workstreams: debug, / refactor, tests, reliability / ● 99.9% crash-free, 4.9/5, #1 on Google Play (Six bullets, each a solid black dot bullet in Arial at x 54.9. Text starts at x 57.8. Line spacing about 1.2 times font size within a bullet. About 9 percentage points of vertical gap between bullets. Bold runs: '~85%' and '99.9%'.)
- logo at x 2.7, y 94.4, w 5.5, h 1.5 [OpenAI Sans, Bold weight, sans-serif; font size 1.48% of slide height; #000000; left aligned]: OpenAI wordmark (Small black wordmark in the bottom-left corner, aligned with the left edge of the image. Estimated box.)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans, Regular weight, sans-serif, normal case; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Sits to the right of the OpenAI wordmark on the same baseline.)
- page number at x 96.3, y 94.4, w 1, h 1.5 [OpenAI Sans, Regular weight, sans-serif, normal case; font size 1.48% of slide height; #000000; right aligned]: 35 (Bottom-right corner, right margin about 2.7% of slide width.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 2.7, y 6.7, w 46.7, h 83.0: 1024 by 1024 pixels
- text at x 53.8, y 13.3, w 25.8, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: Shipping Sora in
- text at x 53.8, y 20.4, w 29.8, h 5.9 [OpenAISans-Medium; font size 5.93; #000000]: 28 days with Codex
- text at x 54.9, y 32.2, w 1.2, h 3.5 [ArialMT; font size 3.46; #000000]: ●
- text at x 57.8, y 32.2, w 22.5, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Lean team of 4 engineers
- text at x 54.9, y 41.4, w 1.2, h 3.5 [ArialMT; font size 3.46; #000000]: ●
- text at x 57.8, y 41.4, w 37.3, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: 18-day to prototype and 10 days to launch
- text at x 54.9, y 50.6, w 1.2, h 3.5 [ArialMT; font size 3.46; #000000]: ●
- text at x 57.8, y 50.6, w 11.6, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Codex wrote
- text at x 69.4, y 50.6, w 5.1, h 3.5 [OpenAISans-Bold; font size 3.46; #000000]: ~85%
- text at x 74.6, y 50.6, w 18.9, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: of the code, humans
- text at x 57.8, y 55.4, w 31.1, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: focused on architecture and review
- text at x 54.9, y 64.6, w 1.2, h 3.5 [ArialMT; font size 3.46; #000000]: ●
- text at x 57.8, y 64.6, w 32.1, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: Parallel Codex workstreams: debug,
- text at x 57.8, y 69.4, w 21.2, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: refactor, tests, reliability
- text at x 54.9, y 78.6, w 1.2, h 3.5 [ArialMT; font size 3.46; #000000]: ●
- text at x 57.8, y 78.6, w 6.0, h 3.5 [OpenAISans-Bold; font size 3.46; #000000]: 99.9%
- text at x 63.8, y 78.6, w 31.8, h 3.5 [OpenAISans-Regular; font size 3.46; #000000]: crash-free, 4.9/5, #1 on Google Play
- text at x 96.3, y 94.4, w 1.0, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: 35
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 36

**Text**

I’ll leave everyone with 3 next steps:
1) Use Codex in a real repo
2) Set up Agents.md
3) Give it real tasks
We have additional sessions
scheduled to dive deeper!
36
Confidential and proprietary.

**Layout**

Two-column split. The left 49.3% of the slide is a full-height blue-violet panel carrying a dark app screenshot that runs off the left edge. The right half is white with a left-aligned text block starting at about 54% of the slide width, vertically centred, with no title bar. The page number sits bottom right at about 96% x and 94% y, and the confidentiality line sits bottom left at about 8% x and 94.5% y, inside the photo panel. Margins are roughly 4% at the right and 6% at the top and bottom.

**Visuals**

Photograph or illustration (left panel): a dark-mode desktop app window titled 'Codex — web-app' with rounded corners and a soft drop shadow. The window is cropped by the left slide edge, so its left portion is cut off. Inside it, a split diff view on the left shows React/Next.js source code (import type React from 'react'; import Link from 'next/link'; import { ArrowRight, Check, Rocket, Shield, Zap } from ...; import { Button } from '@/components/ui/button'; import ThemeToggle; export default function Page() {...; a '<section id="testimonials"' block with an h2 'Loved by builders like you' and a p 'Real teams ship faster and stay co...'). Added lines are shaded green and removed lines red. A right pane shows the Codex agent chat for the prompt 'Add a new testimonial section to my landi...', with a vertical list of steps (Explored files in app; Read page.tsx; Clarifying testimonial request; Exploring testimonial section options; Considering testimonial section changes; Explored files in components; Opening tailwind config; Read tailwind.config.json; Next, I'll add a reusable Testimonials component and swap it into app/page.tsx to keep things tidy; Created Testimonials.tsx +102 -0; Updating app/page.tsx; Edited page.tsx +6 -70; Running Lint and Typecheck; Running prettier...). A summary bar reads '2 files changed +108 -70' with a 'View changes' button. Below is a text input 'Ask for follow-up changes' with '+' and 'Auto' controls, and a footer row with 'Local', 'Agent (full access)' and 'Medium'. A bottom terminal panel lists tabs 'Output', 'Terminal', 'Ports', 'GitLens', 'Azure', with output showing 'Local: http://localhost:3000' and 'Network: http://127.0.2.2:3000' followed by compile timings and 200 status lines. Composition is a landscape window placed in the lower-left, with the top-left and left edges cropped. Lighting is flat, UI-based, with no photographic lighting. Dominant colours are charcoal #1E1E1E, dark grey #2A2A2A, muted red #5A2A2A and green #2A4A2A diff tints, and cool blue-violet from the backdrop. The screenshot is boxed, not full-bleed, and has no overlay on it.

**Design**

Background: Split background: white #FFFFFF on the right half and blue-violet gradient (#E9E6FF to #4B4FE8) on the left panel. Clean white right half with black sans-serif text on a split layout. The left panel uses a soft blue-violet gradient: pale lavender #E9E6FF at upper centre, transitioning to saturated blue #4B4FE8 and violet #5B5CF0 toward the lower left. Body and list text is black #000000, OpenAI Sans Regular, a geometric grotesque sans-serif in regular weight, set at about 4.7% of slide height with no distinct title styling. Accents: none on the text. Numbered list uses 1) 2) 3) markers in the same black type. Line spacing is generous, roughly 1.3 of the size. The screenshot window has rounded corners of about 1.5% of slide width and a soft dark drop shadow. No gradients on text. The page number and confidentiality line are very small (about 1.5% of slide height) in black.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- shape at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide white background fill (Covers the entire slide with no outline; sits at the back.)
- photo at x 0, y 0, w 100, h 100 [#FFFFFF]: Full-slide background image layer, 1920 by 1080 pixels, visible only where the left panel does not cover it (Full-bleed, not cropped; the right half reads as white. No overlay visible.)
- photo at x 0, y 0, w 49.3, h 100 [#5B5CF0]: Blue-violet gradient backdrop panel, 769 by 684 pixels, with no text (Boxed to the left half, full height, butted against the white right half with a hard vertical edge. Gradient runs from pale lavender #E9E6FF at upper centre to saturated blue #4B4FE8 at lower left.)
- photo at x 0, y 14, w 48.9, h 76.4 [#1E1E1E]: Dark-mode app window screenshot 'Codex — web-app', 1000 by 797 pixels, showing a code diff, a Codex agent chat panel and a terminal (Rounded window corners of about 1.5% of slide width, soft dark drop shadow, cropped by the left slide edge. Overlaps the blue gradient backdrop. No overlay or opacity change.)
- title at x 54, y 23.9, w 43, h 4.7 [OpenAI Sans, Regular weight, sans-serif, normal case, no letter-spacing; font size 4.69% of slide height; #000000; left aligned]: I’ll leave everyone with 3 next steps: (Single line, first paragraph of the text block.)
- body at x 54.7, y 40.2, w 33.2, h 4.7 [OpenAI Sans, Regular weight, sans-serif, normal case, no letter-spacing; font size 4.69% of slide height; #000000; left aligned]: 1) Use Codex in a real repo (Numbered list item 1 with a parenthesis marker; sits after a blank line below the intro.)
- body at x 54.1, y 46.4, w 26.4, h 4.7 [OpenAI Sans, Regular weight, sans-serif, normal case, no letter-spacing; font size 4.69% of slide height; #000000; left aligned]: 2) Set up Agents.md (Numbered list item 2 with a parenthesis marker; line spacing matches item 1.)
- body at x 54.1, y 52.6, w 24.7, h 4.7 [OpenAI Sans, Regular weight, sans-serif, normal case, no letter-spacing; font size 4.69% of slide height; #000000; left aligned]: 3) Give it real tasks (Numbered list item 3 with a parenthesis marker; ends the list.)
- body at x 54, y 69, w 34.4, h 4.7 [OpenAI Sans, Regular weight, sans-serif, normal case, no letter-spacing; font size 4.69% of slide height; #000000; left aligned]: We have additional sessions / scheduled to dive deeper! (Two lines of body text; the break after 'sessions' is a line break. The second line is at y 75.2 and w 30.9.)
- body at x 54, y 75.2, w 30.9, h 4.7 [OpenAI Sans, Regular weight, sans-serif, normal case, no letter-spacing; font size 4.69% of slide height; #000000; left aligned]: scheduled to dive deeper! (Second line of the paragraph above, same style.)
- page number at x 96.3, y 94.4, w 1, h 1.5 [OpenAI Sans, Regular weight, sans-serif, normal case; font size 1.48% of slide height; #000000; right aligned]: 36 (Bottom-right slide number.)
- footer at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAI Sans, Regular weight, sans-serif, normal case; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Bottom-left confidentiality line; it sits over the blue-violet panel and is barely legible at this size.)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 0.0, y 0.0, w 100.0, h 100.0: #FFFFFF (covers the slide)
- image at x 0.0, y 0.0, w 100.0, h 100.0: 1920 by 1080 pixels (covers the slide)
- image at x 0.0, y 0.0, w 49.3, h 100.0: 769 by 684 pixels
- image at x 0.0, y 14.0, w 48.9, h 76.4: 1000 by 797 pixels
- text at x 54.0, y 23.9, w 43.0, h 4.7 [OpenAISans-Regular; font size 4.69; #000000]: I’ll leave everyone with 3 next steps:
- text at x 54.7, y 40.2, w 33.2, h 4.7 [OpenAISans-Regular; font size 4.69; #000000]: 1) Use Codex in a real repo
- text at x 54.1, y 46.4, w 26.4, h 4.7 [OpenAISans-Regular; font size 4.69; #000000]: 2) Set up Agents.md
- text at x 54.1, y 52.6, w 24.7, h 4.7 [OpenAISans-Regular; font size 4.69; #000000]: 3) Give it real tasks
- text at x 54.0, y 69.0, w 34.4, h 4.7 [OpenAISans-Regular; font size 4.69; #000000]: We have additional sessions
- text at x 54.0, y 75.2, w 30.9, h 4.7 [OpenAISans-Regular; font size 4.69; #000000]: scheduled to dive deeper!
- text at x 96.3, y 94.4, w 1.0, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: 36
- text at x 8.2, y 94.5, w 10.9, h 1.5 [OpenAISans-Regular; font size 1.48; #000000]: Conﬁdential and proprietary.

### Slide 37

**Text**

Resources
Codex
Documentation
Codex
Cookbooks
Codex
Changelog
developerss.openai.com/codex
cookbook.openai.com/topic/codex
developers.openai.com/codex/changelog
37
Confidential and proprietary.

**Layout**

Three equal white cards sit in a row across the lower two-thirds of the slide, with a single title at top left above them. The title is left-aligned at about x 5.3% and y 9.0%. The cards are at x 3.1%, 35.4% and 67.8%, each 29.1% wide and 51.7% high, starting at y 23.9%, with equal gutters of about 3.2%. Each card holds an underlined link title, an italic URL line and a centred QR code. The slide number sits bottom right at x 96.4%, y 94.4%. A small confidentiality line sits bottom left at x 8.2%, y 94.5%.

**Visuals**

Three QR codes, one in each card, all black modules on white. Card 1 QR at x 13.2%, y 57.2%, w 8.9%, h 15.8%, 121 by 122 pixels. Card 2 QR at x 45.6%, y 57.1%, w 8.9%, h 16.1%, 116 by 118 pixels. Card 3 QR at x 78.0%, y 57.1%, w 8.9%, h 15.8%, 115 by 115 pixels. Each QR is square, unframed and sits on white, pointing to the URL printed above it. The background is a soft blue-to-lavender gradient photograph with cloud-like streaks, lighter pale lavender top centre and top right, deeper blue and violet at the right and bottom right, covering the full slide. No other illustrations, logos or icons.

**Design**

Background is a full-bleed blue and lavender gradient with soft cloud texture. Card fill is pure white #FFFFFF with square-cornered edges, a small corner radius of about 0.5% of the slide width and no visible border or shadow. Titles and body text are black #000000. Typeface is OpenAI Sans, a geometric grotesque sans-serif. Title text is regular weight, about 5.4% of slide height. Card titles are regular weight, about 5.4% of slide height, underlined. URLs are italic, about 2.2% of slide height. Slide number and confidentiality line are Helvetica Neue regular at about 1.5% of slide height. No accent colours.

**Elements and positions** (back to front; x, y, width, height as % of the slide; type, colour and alignment in brackets)

- photo at x 0, y 0, w 100, h 100: Full-bleed blue and lavender gradient background image with cloud texture, covering the slide (Image 1600 by 900 pixels (also a 1920 by 1080 version covers the slide); not cropped; no overlay; behind all other elements)
- shape at x 3.1, y 23.9, w 29.1, h 51.7 [#FFFFFF]: White rounded rectangle card 1 (Corner radius about 0.5% of slide width; no border; no shadow; opacity 100%)
- shape at x 35.4, y 23.9, w 29.1, h 51.7 [#FFFFFF]: White rounded rectangle card 2 (Corner radius about 0.5% of slide width; no border; no shadow; opacity 100%)
- shape at x 67.8, y 23.9, w 29.1, h 51.7 [#FFFFFF]: White rounded rectangle card 3 (Corner radius about 0.5% of slide width; no border; no shadow; opacity 100%)
- title at x 5.3, y 9, w 14.5, h 5.4 [OpenAI Sans, Regular; font size 5.43% of slide height; #000000; left aligned]: Resources
- title at x 5.7, y 29.4, w 21.7, h 11.9 [OpenAI Sans, Regular, underlined; font size 5.43% of slide height; #000000; left aligned]: Codex / Documentation (Underlined hyperlink text; line 1 'Codex' at y 29.4%, line 2 'Documentation' at y 36.0%)
- title at x 38.1, y 29.4, w 19.7, h 11.9 [OpenAI Sans, Regular, underlined; font size 5.43% of slide height; #000000; left aligned]: Codex / Cookbooks (Underlined hyperlink text; line 1 'Codex' at y 29.4%, line 2 'Cookbooks' at y 36.0%)
- title at x 70.4, y 29.4, w 23.4, h 11.9 [OpenAI Sans, Regular, underlined; font size 5.43% of slide height; #000000; left aligned]: Codex / Changelog (Underlined hyperlink text; line 1 'Codex' at y 29.4%, line 2 'Changelog' at y 36.0%)
- body at x 5.7, y 48.5, w 17, h 2.2 [OpenAI Sans, Italic; font size 2.22% of slide height; #000000; left aligned]: developers.openai.com/codex
- body at x 38.1, y 48.5, w 19.7, h 2.2 [OpenAI Sans, Italic; font size 2.22% of slide height; #000000; left aligned]: cookbook.openai.com/topic/codex
- body at x 70.4, y 48.5, w 23.4, h 2.2 [OpenAI Sans, Italic; font size 2.22% of slide height; #000000; left aligned]: developers.openai.com/codex/changelog
- icon at x 13.2, y 57.2, w 8.9, h 15.8 [#000000]: QR code linking to developers.openai.com/codex (Black modules on white, square, 121 by 122 pixels, no frame)
- icon at x 45.6, y 57.1, w 8.9, h 16.1 [#000000]: QR code linking to cookbook.openai.com/topic/codex (Black modules on white, square, 116 by 118 pixels, no frame)
- icon at x 78, y 57.1, w 8.9, h 15.8 [#000000]: QR code linking to developers.openai.com/codex/changelog (Black modules on white, square, 115 by 115 pixels, no frame)
- page number at x 96.4, y 94.4, w 0.9, h 1.5 [Helvetica Neue, Regular; font size 1.48% of slide height; #000000; right aligned]: 37
- footer at x 8.2, y 94.5, w 10.6, h 1.5 [Helvetica Neue, Regular; font size 1.48% of slide height; #000000; left aligned]: Confidential and proprietary. (Listed in the file measurements at this position but not visibly rendered on the slide image)

**Measured from the file** (exact, read from the PDF rather than estimated; x, y, w, h as % of the slide; font size as % of the slide height)

- fill at x 3.1, y 23.9, w 29.1, h 51.7: #FFFFFF
- fill at x 35.4, y 23.9, w 29.1, h 51.7: #FFFFFF
- fill at x 67.8, y 23.9, w 29.1, h 51.7: #FFFFFF
- image at x 0.0, y 0.0, w 100.0, h 100.0: 1920 by 1080 pixels (covers the slide)
- image at x 0.0, y 0.0, w 100.0, h 100.0: 1600 by 900 pixels (covers the slide)
- image at x 45.6, y 57.1, w 8.9, h 16.1: 116 by 118 pixels
- image at x 78.0, y 57.1, w 8.9, h 15.8: 115 by 115 pixels
- image at x 13.2, y 57.2, w 8.9, h 15.8: 121 by 122 pixels
- text at x 5.3, y 9.0, w 14.5, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Resources
- text at x 5.7, y 29.4, w 8.9, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Codex
- text at x 38.1, y 29.4, w 8.9, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Codex
- text at x 70.4, y 29.4, w 8.9, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Codex
- text at x 5.7, y 36.0, w 21.7, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Documentation
- text at x 38.1, y 36.0, w 15.7, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Cookbooks
- text at x 70.4, y 36.0, w 15.2, h 5.4 [OpenAISans-Regular; font size 5.43; #000000]: Changelog
- text at x 5.7, y 48.5, w 17.0, h 2.2 [OpenAISans-Italic; font size 2.22; #000000]: developers.openai.com/codex
- text at x 38.1, y 48.5, w 19.7, h 2.2 [OpenAISans-Italic; font size 2.22; #000000]: cookbook.openai.com/topic/codex
- text at x 70.4, y 48.5, w 23.4, h 2.2 [OpenAISans-Italic; font size 2.22; #000000]: developers.openai.com/codex/changelog
- text at x 96.4, y 94.4, w 0.9, h 1.5 [HelveticaNeue; font size 1.48; #000000]: 37
- text at x 8.2, y 94.5, w 10.6, h 1.5 [HelveticaNeue; font size 1.48; #000000]: Conﬁdential and proprietary.
