Rohan Varma

I’m joining OpenAI Codex to work on the future of agentic development! At Cursor, I got to see the shift from autocomplete to agents. The next step isn’t a better IDE. It’s an Agent Development Environment (ADE): systems and tools for orchestrating agents, reasoning over their outputs, and making them autonomous enough to reliably complete ambitious work. After chatting with @embirico and @thsottiaux, it was clear that Codex is the best place to realize this vision. The team has consistently shipped SOTA models for agentic coding (check out gpt-5.3-codex) and I’m pumped for the future that the new Codex App points to. What I’m most excited about is the broader mission: accelerating the knowledge work economy. All agents are coding agents, and we’re already seeing Codex used across every job function within organizations. I’m extremely grateful for my time at Cursor, working with the incredible team, and I’m proud of what we built together. I’m excited to take an even bigger swing with Codex. If you’re curious to get a glimpse of where we are headed, download the Codex App! If you want to work on this mission, please apply or reach out - we are hiring across all functions! You can just build things.

@dunkhippo33 I think you’d like GPT-5.5 if you check it out :)

I don't know if it's just me, but I'm finding that the latest Opus model is not that great, and in fact, I still get a lot of benefit out of Sonnet.

@fortelabs Codex

Can anyone recommend a dead simple, easy to use notetaking app based on markdown files? Obsidian is far too complicated for me. I'm looking for Apple Notes, but with markdown storage

@rawBit_io Was this in codex or chat? Would be curious if it could genetically decide on improvements and then invoke image gen as needed to express any ideas

@TheRohanVarma I didn't specify the exact problem, it was more like a search for better alternative UI. I provided the left image and this is screenshot of my prompt

Our 2.0 image model is so good at making screens and vision mocks. Something about AI generated images of digital surfaces feels very “right” to me. Internally, I’ve started seeing tons of product ideas shared and brought to life via image generation rather than prototyping — pretty cool! Using it in Codex makes Codex a full stack design engineer that can build beautiful interfaces. Would love to hear what creative ways y’all are leveraging it for engineering 👀 (This image was one-shot generated with “Make a screen in the Codex App on a Mac desktop that is an AI code review view for PRs.”)

@rawBit_io Curious in what ways you were hoping to improve the UX in the left image? Would be interested to see if, when specifying the problems to be solved for in the redesign, imagegen is able to account for that

@TheRohanVarma In my experience UI from image generator makes little sense, even with whole context provided. On the left is my original UI, on the right is 'improved' UI from the image 2.0 As context I provided link to GitHub repo github.com/rawBit-io/rawb…

@nicdunz You’re going to be s u p e r excited for what we have in store Also openai.com/index/introduc…

wtf does the chatgpt team even do? theres so many outdated things waiting on updates. memory, personalization, voice, tasks, etc. etc. etc. stop abandoning chatgpt!!!!!!

@yacineMTB Should be possible with a plugin

codex please change my diaper because i'm a big fat baby

Translation: you can get more work done with GPT-5.5, compared to other frontier models.

Period

@svpino Have you met @steipete

I don’t believe people who say they are running “12 parallel coding agents”. Either they are lying for clicks, or I’m a complete retard who can barely keep up with a single Claude instance.

@ctatedev The turns have tabled it seems 👀

GPT-5.5 for code Opus 4.7 for review

It’s actually quite hard to have agents always working on your behalf, productively. Last week I chatted with some of the top OpenAI employees on our internal Codex token leaderboard (you don’t want to know how many tokens they are using 😅) and they all shared some pretty cool workflows and tools they used to have agents always working for them. My favorite example was one engineer who forked the Codex App and built entire systems into it that enable him to have agent swarms running iteratively on tasks (some cool stuff coming here later 👀). In some sense, the goal of the App is to make it easy to maximally leverage agents, through sync and async techniques. What do you do to keep agents usefully running for you? What don’t we support today that would make this easier?

@ajambrosino @simpsoka Big if big

@simpsoka big if true

It’s true, I am codexing things right now

@maddiedreese Codex

After Google won search, “google” became a verb. It’ll be interesting to see which AI lab turns its product into a verb first!

@ajambrosino Great day to deliver shareholder value

a great day to spearhead some initiatives

In the last week we launched: - GPT-5.5 - Codex App artifacts (sheets, slides, etc) - Codex Browser and Computer Use - Workspace Agents Using these features for the last 2-3 weeks have *actually* changed the way I work. Would recommend folks play around with all the new tools and really try experimenting with your workflows!

@sama 🫡 @thsottiaux get the button ready

@rohanvarma how about a rate limit reset in a week or so?

For clarity, we are running a small test for ~100% of Codex users where we: - make our best models available to all users - make Codex available to all plans, free and paid Claude Code users aren't affected :)

Knowing Codex is better for complex tasks makes me bullish on @chamath and 8090 He points out that Codex is much better for meaty, large problems, while Claude feels good for more day to day work. We’re improving model personality so Codex feels even better for pair programming. But we think pair programming with agents is a local optimization. The longer term pattern is delegation. Most people just aren’t asking agents to do ambitious enough work. The people who are delegating use Codex.

“The world is healing” Thankful that we don’t have to keep using a terminal to do powerful agentic things