How Does Codex Improve Developer Workflows?

TL;DR

Codex now combines GPT-5.6 Sol, Ultra mode, automatic subagents, faster computer and browser use, inline editing, deployment, task coordination, mobile access, and pull request management in one ChatGPT workspace. Developers can move from difficult project planning to parallel implementation, visual testing, full-stack publishing, and code review while preserving context across conversations, branches, devices, and related tasks.

Transcript

This is a big moment for Codex and I'm really excited to show you what's new for developers. Today, more than 6 million of you use Codex every week. Along the way, you've shared feedback, shown us what you're building, and you set an incredibly high bar. You've asked Codex to ship ambitious projects and help you stay organized beyond the code. You'... Read More

Key Insights

• Codex and ChatGPT are now combined in one app, with Codex occupying a dedicated space beside the new work agent. Developers retain the established Codex experience while gaining a direct path from ChatGPT conversations and deep research into implementation work without manually copying context.
• GPT-5.6 Sol is the new frontier model available to everyone, and it is designed to work through difficult problems for extended periods. Ultra mode supplies a much larger reasoning budget for especially challenging projects, although using it consumes token limits more quickly.
• Automatic subagent coordination lets Codex divide one goal into multiple parallel workstreams without requiring the developer to request that split. Developers can follow an individual subagent or open a panel showing the entire team, making concurrent work visible from a single coordinating thread.
• Appshots provide both the visible screen and relevant app context to Codex. In the demonstrated iPhone workflow, Codex navigates every Simulator screen and captures App Store screenshots in English and French while operating in the background, leaving the developer free to continue using the computer.
• The in-app browser supports direct testing and visual feedback for applications created by Codex. Developers can play an application, point to a specific interface area, annotate the requested change, and give Codex the visual context needed to modify that exact part.
• Inline diff editing supports both manual and delegated code changes. A developer who knows the precise edit can change a line directly, while someone who knows only the desired behavior can describe that outcome and ask Codex to determine the necessary implementation.
• Sites publishes Codex-built web applications with hosting, authentication, a persistent database, and file storage included. Previously introduced for Enterprise teams to share internal applications, Sites is now available to everyone and works with any web application built using Codex.
• Codex supports connected workflows across projects, devices, and pull requests. It can organize separate tasks and isolated worktrees, surface open pull requests for task branches, incorporate comments or failed checks, and let developers review, mark ready, and merge changes without visiting GitHub.com.
• More videos from this channel:
  • How to design frontends with OpenAI Codex multimodal
  • What is Codex in ChatGPT and how does it work?
  • How to use Codex CLI with GPT-5 Codex
• More from this channel: OpenAI

Questions & Answers

Q: How does GPT-5.6 Sol improve Codex for developers?

GPT-5.6 Sol is presented as a new frontier model that is available to everyone and capable of working through difficult problems for a long time. It also improves computer and browser use, making those interactions faster. When developers need more reasoning capacity, they can select Ultra mode and combine it with the Goal command for demanding projects they have postponed.

Q: What is Ultra mode in Codex?

Ultra mode gives GPT-5.6 Sol a much larger reasoning budget for especially challenging work. The demonstration suggests using it with the Goal command when assigning ambitious projects that require sustained problem solving. The tradeoff stated in the presentation is that Ultra consumes token limits faster, so developers should reserve it for work that benefits from additional reasoning capacity.

Q: How does Codex coordinate parallel work with subagents?

Codex can automatically split a goal across multiple subagents without requiring the developer to ask for parallelization. The developer can follow each subagent individually or open a panel that displays the whole team working together. A coordinating thread can also create separate tasks and isolated worktrees, monitor their progress, and pin the most critical items.

Q: What is an appshot in Codex?

An appshot is described as a smarter screenshot because it provides Codex with both the screen and the app context. On a Mac, the demonstrated workflow captures one by pressing Cmd-Cmd. Codex can then operate the Simulator, navigate through screens, and capture App Store screenshots in multiple requested languages while working in the background.

Q: How can developers edit applications visually in Codex?

Developers can open an application in the in-app browser, interact with it, and inspect what Codex created. They can point and click to annotate the exact interface element they want changed, giving Codex visual context for the request. After the modification, they can inspect the diff and make precise inline code edits or describe the desired behavior.

Q: What infrastructure does Codex Sites provide?

Sites can publish any web application built with Codex and includes hosting, authentication, a persistent database, and file storage. A developer can request publication directly from the Codex workflow, allowing an idea to become a deployed full-stack application without separately managing the demonstrated infrastructure and hosting components. Sites was introduced for Enterprise before becoming available to everyone.

Q: How can developers manage Codex tasks from a phone?

ChatGPT on mobile lets developers create, search, open, and manage Codex tasks directly from a conversation. They can review changes by filtering staged changes, unstaged changes, branches, or branch comparisons. They can also connect directly to an SSH host, such as a development box, and start new tasks while away from their desktop.

Q: How does Codex handle pull request workflows?

Codex tracks the branch associated with each task and automatically surfaces an open pull request in the summary pane when one exists. Developers can follow progress, resolve issues, respond to comments, and ask Codex to fix failed checks with full task context. The Pull Requests tab also supports inline comments, complete reviews, readiness updates, and merging without visiting GitHub.com.

Summary & Key Takeaways

• Codex now has a dedicated space inside ChatGPT alongside the new work agent. GPT-5.6 Sol is available to everyone, while Ultra mode gives it a larger reasoning budget for difficult projects. Codex can also divide a goal among subagents automatically, enabling several coordinated pieces of work to progress in parallel.

• Computer and browser use let Codex interact with apps and websites while developers continue using their computers. Appshots combine screen information with app context, the in-app browser supports visual annotation and authenticated applications with passkeys, and the diff view allows developers to edit code directly or request behavior-based changes from Codex.

• Sites can publish Codex-built web applications with hosting, authentication, persistent databases, and file storage included. Codex also coordinates tasks, isolated worktrees, branches, and pull requests across desktop and mobile. Developers can carry ChatGPT conversation context into Codex, connect to SSH hosts, review changes, address feedback, and merge pull requests.