Video summary
Codex just got better for developers
Main summary
Key takeaways
Summary of technological concepts & product updates (Codex)
Codex integrated into ChatGPT
- Codex is integrated into the ChatGPT app
- It appears in a dedicated space next to a new work agent.
- The experience is meant to remain developer-focused, similar to the Codex product.
Frequent iteration / rapid releases
- The team reports 150+ updates in the past two months.
- Improvements continue on a weekly cadence.
Model and coding/task delegation
New frontier model: “GPT-5.6 Sol”
- Positioned as a model available to everyone.
- Claims it can work on hard problems for a long time.
- Ultra mode is available with GPT-5.6:
- Uses a larger reasoning budget.
- Tradeoff: Ultra consumes token limits faster.
Parallelized execution via subagents
- Codex can automatically split work across subagents (no manual coordination needed).
- Users can:
- View individual subagents
- Or view a panel showing the full parallel “team” working toward a goal.
Browser/computer automation (UI capture and test-like flows)
Faster computer/browser use with GPT-5.6
- Demo focus: creating an “appshot” from a simulator (described as a smarter screenshot).
- The appshot includes:
- Screen pixels
- App context
- Codex can then use the simulator itself to navigate UI screens.
- Example includes capturing App Store screenshots in English and French.
- Key claim: automation happens without the user controlling the cursor, while the user can continue working.
Visual editing and code editing in the app/browser
In-app browser support + interactive editing
- Codex can generate and run a simple web app (e.g., a “paper plane game”) inside the in-app browser.
- Users can point-and-click annotate to request changes using visual context.
Browser enhancements for real apps
- Improved support for web apps requiring authentication, including passkeys.
Diff view + inline precision edits
- Users can switch to diff view.
- Codex supports precise inline edits beyond inspection.
- Two common workflows:
- Make changes directly if you know the code change
- Ask Codex to infer changes if you know the intended behavior
Publishing/deploying apps
Sites for everyone (full-stack hosting)
- Previously introduced for Enterprise, now available more broadly.
- “Publish to Sites” includes:
- Hosting
- Authentication
- Persistent database
- File storage
- Goal: move from idea to a deployed full-stack app without managing infrastructure.
Project/task organization and coordination
Codex helps manage its own tasks
- Example workflow:
- Find bugs from Linear
- Create separate tasks and isolated worktrees per bug
- Pin the most critical work
- A coordinating view shows what is kicked off and what is moving.
Task cross-references
- Tasks can reference one another to link related work.
Bring context from ChatGPT
- If work begins in ChatGPT (e.g., via deep research), users can @mention the conversation to transfer context into Codex—avoiding copy/paste.
Mobile upgrades (ChatGPT on mobile)
Task management from conversations
- Create, search, open, and manage tasks directly from a mobile conversation.
Better review workflow on mobile
- Filter review changes by:
- Staged
- Unstaged
- Branch
- Compare branches
Remote development connectivity
- Connect directly to an SSH host (e.g., a dev box) to start new tasks from a phone.
PR (Pull Request) workflow inside Codex/ChatGPT
PRs without leaving the app
- Codex understands which branch each task relates to.
- When a branch has an open PR, the summary pane surfaces it automatically.
Fixing PR issues with full context
- If checks fail or comments appear, users can ask Codex to fix them in place using full context.
New PRs tab
- Explore code changes, leave inline comments, perform full reviews.
- Mark PRs as ready and merge directly.
- Emphasis: keep the workflow inside Codex rather than going to GitHub.com.
Review/guides/tutorial references
- This is a product demo rather than a step-by-step written tutorial.
- It includes practical “how it works” demos for:
- Goal + Ultra mode delegation
- Appshot + simulator automation
- Browser-based visual annotation edits
- Diff view + inline code edits
- “Publish to Sites” deployment
- Task coordination tied to Linear and worktrees
- Mobile task/review/SSH workflows
- In-app PR review and issue remediation
Main speakers/sources
- Primary speaker: the presenter (the name is not explicitly provided in the extracted text; referenced in subtitles only as “Let me show you…”).
- Source/org referenced: OpenAI (and Codex, ChatGPT, Build Week, Linear, GitHub/GitHub.com as applicable in descriptions).
Rate this summary
Your feedback will help improve summaries.
Improve this summary
Reprocess with a stronger model when the summary feels incomplete or inaccurate.
Translate summary in another language
Ask questions to this video
Chat for follow-up questions, clarifications, and source-backed answers.