We're adding support for AGENTS.md to Claude Code. Starting today in version 2.1.277, if there is no CLAUDE.md in a folder, Claude will check for and use AGENTS.md. You can toggle this behavior in /config.

OK nevermind, the team and Astra did a good job and our systems are more scalable than we anticipated. Astra is now rolled out to all Plus and Business users too. Hope you have a blast and let us know how it goes!

wrote down some of the design thinking behind Grok Bot. persistent roles, clear state, scoped context, coordinated teams — an interface designed to move you from operating AI to delegating work.

Build a plugin once and use it across compatible agent clients. Introducing Agent Plugins, an open standard developed with @awsdevelopers, @cursor_ai, @github, @code, and @vercel that packages Agent Skills and supports MCP server configurations in a shared format.

Introducing GPT-Live, a new generation of voice models for natural human-AI interaction. Rolling out in ChatGPT starting today. You’ll want to turn the sound on for this one.

Applications to attend OpenAI DevDay 2026 are open. Join us in San Francisco on September 29 to: • Tinker with what’s new • Swap build notes with other builders • Go deep in technical sessions • Bring your sharpest questions Apply by July 10:

Backstage at @aiDotEngineer! The energy here is incredible. Excited to kick off the opening keynote with @embirico in a few minutes. We might even be joined by a special guest… 👀

Show Codex a workflow once. Reuse it as a skill. Record & Replay lets you show Codex a recurring task, like filing an expense report or submitting a time-off request. Codex turns that demo into an inspectable, editable skill. You control when recording starts and stops.

Career update: I’ve joined @OpenAI to lead Cyber as Head of Product for Cyber, where I’ll be leading OpenAI’s efforts to bring frontier AI capabilities to cyber defenders and help make software safer and more resilient by design. I’ll be teaming up with @clintgibler, an expert engineering and security research leader, who is announcing he is joining today as well. What is now possible with frontier AI models represents a genuine step change for defenders. I’m particularly excited about the potential to improve how we: • Analyze code and discover vulnerabilities. Models can reason across large and unfamiliar codebases, identify subtle weaknesses, focus attention on realistic attack paths, and help teams move faster from discovery to validated remediation. • Investigate incidents and determine root causes. Security teams spend enormous time connecting fragmented signals across code, infrastructure, identities, endpoints, and applications. AI can help bring those events together, form and test hypotheses, and surface root causes faster. • Orchestrate security work. The opportunity is not only to generate recommendations, but to help defenders safely execute work: validating findings, testing patches, improving detections, gathering evidence, and coordinating remediation. • Make enterprise AI agents safe and controllable. As agents take on more meaningful work inside organizations, they need secure harnesses: scoped access, isolation, monitoring, verification, auditability, and clear human control. Security must be built into how agents operate from the beginning. Our work at OpenAI starts from a simple but ambitious premise: the next generation of cyber defense should be integrated into how software is built, not only finding and patching vulnerabilities, but making systems resilient from the start. With current model capabilities such as GPT-5.5-Cyber, alongside Codex as an agentic harness, we can give defenders more powerful tools while pairing those capabilities with appropriate verification, safeguards, accountability, and control. Just as importantly, we need to distribute these defensive advantages broadly. Powerful security capabilities should not be limited to the largest organizations with the biggest teams. They should help developers, security practitioners, public institutions, and businesses of every size build and operate more secure systems. And no single company can accomplish this alone. Security has always been a community effort. I’m looking forward to working closely with leaders and innovators across the cybersecurity ecosystem: vendors, researchers, practitioners, governments, and technology partners. AI will create new security challenges, but it also gives us a remarkable opportunity to strengthen the defensive ecosystem. I’m grateful to the OpenAI team for the opportunity and excited to get started on a mission that matters deeply to me.

📣 Open call to agent builders: Let's read agent skills from `.agents/skills`, so people don't have to manage separate folders per agent. Today we pulled the trigger for Codex to read `.agents/skills`. Goal is to deprecate `.codex/skills`. Pls like/tag/RT for momentum.

grug is probably the first Apple Design Award winner built by two designers using Codex to write the code. We didn't prompt "build me an award-winning app, make no mistakes.” It did not wake up one morning and decide the world needed grug. We did. We wanted to build it and AI was the tool that helped us take it all the way there. Ever since we started building with AI a few years ago, we got bolder. You stop killing ideas just because they sound too hard to build. You get weirder when trying things becomes cheap enough to be silly again. You start following the strange little thought further than you normally would. That is how grug happened. AI can write your code. It can help you move fast. It can make impossible things feel possible. But it cannot care. It cannot make your app memorable. It cannot make your app feel like it has a soul. You have to bring the taste. You have to stay incredibly close. You have to take small steps, make thousands of tiny decisions, throw away good-enough work to get to your best work, and protect the thing that made the idea worth building in the first place. grug would not have been this memorable if we didn't have Codex to go all-in on all the crazy ideas we had. If we had not decided to make the whole thing hand-drawn. If we had not spent days and nights obsessing over every detail, every animation, every interaction, every tiny bit of weirdness. That is the difference between slop and something with a soul. And I think that is why this award means so much to us. Not because what we were able to do using AI to build grug, but because Apple recognized the care we put inside all the weirdness. For the past 15 years, so many of our ideas stopped at the mockup. They were too weird, too small, too hard to explain, too expensive to build, too unlikely to survive a meeting. Now designers like us can build the fun little things. Designers like us can ship the crazy ideas. Designers like us can make unreasonable little things and see if the world cares. There are plenty of people who do not get grug. That is totally fine. We do not try to build for everyone, because when you do you end up building for no-one. The people who get grug really get it. And if it makes their morning feel a little lighter, that is all that matters to us. Make the weird thing. The right people will find it. grug no wait for permission. grug back sun rise.

OpenAI frontier models and Codex are now generally available on AWS, giving enterprises a new way to build on Amazon Bedrock with OpenAI through the security, compliance, and governance workflows they already use. This is also the beginning of a broader expansion of OpenAI capabilities on AWS, including future availability for cybersecurity capabilities like Daybreak.

Codex now supports more of the Windows developer loop. With Computer use on Windows, Codex can test apps, debug flows, and review work where your project context lives. Codex in the ChatGPT mobile app lets you connect to Windows machines and keep steering from your phone.

codex power-user best practices: - a few different "pulse" threads that run every morning to check the status of stuff i care about, e.g. proof metrics, all company meetings, etc - a "log" thread for ongoing, everyday activity i want to track. for me that's my "writing log"—want to track what im writing about, and working on from piece to piece - "inbox" thread which gathers all of my emails right now, but eventually will pull in all of the main important things from each "pulse" thread - "router" thread which knows about all of the other threads, and is also hooked up to my email to push emails into each thread appropriately

price performance, the more im at openai the more impressed I am with our training and inference teams here I'm reinterpreting @serenaa_ge's great data really want to know where 4.8 fits into the picture...