Brent Traut

@simpsoka Also, I’ll be joining you on the team in a couple weeks. Looking forward to meeting you all, and to subsequently ship my own “first contribution” +tweet!

@simpsoka Love this! This was definitely a feature I was missing from Conductor. I don’t suppose there’s any way to associate port use with a specific worktree, is there?

We are truly shipping constantly on the Codex team. It doesn't feel like releases as much as a steady stream of things getting better. I landed my first 'larger' PR this week, a small QOL update to the in app browser that replaces the blank empty state. It now discovers local servers from chats and scans common localhost ports. Lemme know what you think!

The new Codex sidebar pairs really well with Every's Compound Engineering 3 skills. It's really clear what design/planning docs were created, and the ce:work skill generates really clear TODOs for Codex to display.

A new feature sneaked in the Codex app’s latest update. You can now do /side (or use the ... menu) to spawn a side chat! Useful when you're deep in a thread and want to have a side question in the current context!

@bridgemindai Price per token is a really lousy metric. Models can chatter away, think forever, and never actually accomplish anything useful. It's difficult to find the right benchmark to objectively, but you really need to measure cost per solution instead. GPT 5.5 is a bargain there!

GPT 5.5 is now the most expensive frontier model in the world. $11.30 per million tokens. Claude Opus 4.7: $10. DeepSeek V4 Pro: $2.20. Kimi K2.6: $1.70. A few months ago Sam said Anthropic was the expensive product for rich people. Today Anthropic is 13% cheaper than OpenAI's flagship. The expensive product is yours now, Sam.

@thsottiaux Now you guys are just showing off.

Don't just reset Codex rate limits for fun, it costs money. Don't just reset Codex rate limits for fun, it costs money. ... but the vibes are good ... I have reset Codex rate limits for ALL paid plans to celebrate a good week and allow everyone to build more with GPT-5.5. Enjoy

@proxy_vector 💯 This has been the case for a while, but it was so clunky with existing tools that it required a ton of extra tooling, prompting, patience, and good luck to get it working.

@btraut The big shift is that browser/app automation stops being a side quest. When the agent can see state, click through UI, and still edit code, debugging starts to feel much closer to real operator work.

The new browser use and computer use inside of Codex are absurdly good. It's a step function improvement to the existing MCP/CLI-based browser use tools and it's not just constrained to the browser. Codex team cooked on this one.

@m13v_ I think you're misunderstanding. Codex literally drives your computer. Your own browser has no idea it's being driven by automation.

@btraut curious how it handles sites that sniff the webdriver flag or fingerprint the profile, that's where the 'just use your real chrome' setup kept breaking for us

@LLMJunky @OpenAI @OpenAIDevs Wearing my hoodie as we speak!

Swag. Thanks @OpenAI @OpenAIDevs! This sweater is perfect 👌 Yall are the best.

@m13v_ Sure, so tell it to use your actual browser and it does that too.

@btraut hit the same wall with codex's internal browser, made it two hops on an auth flow before turnstile killed it. attaching to a real chrome session bypassed it first try. headless fingerprint is the actual ceiling, not the agent loop.

@m13v_ "yours"? I'm talking about Codex (OpenAI) and its internal browser use tools. And yes, I'm also using worktrees, but that's a separate conversation.

@btraut haven't tried yours, but 3 claude code agents in parallel kept fighting over cmd-tab and clipboard until each got its own worktree. multi cursor without those clashes is the unlock

@m13v_ Did you actually use the app? It drives your actual machine. Uses your apps. You can literally see its cursors (yes multiple) as it controls multiple apps at a time.

@btraut the step function holds inside their cloud VM. the moment you need an agent driving your actual machine with real auth state and native apps, local MCP servers are still the only path.

@proxy_vector 💯 This was already the case, but the existing tools for it (including ones that I made on my own) were really clunky. I'm really impressed with the speed of Codex's Browser Use.

@btraut The big unlock is that browser use turns "show me it works" into part of the development loop. Once an agent can click through the product, see the broken state, patch it, and verify again, it stops feeling like codegen and starts feeling like a teammate with eyes.

@thsottiaux - I'm constantly folding open code views in the thread as the agent is working, and they're often folded back shut automatically. I want more code visibility and I also want it to respect state through updates. - The new sound is kinda mid.

@thsottiaux More! - Copy/pasting text usually results in all sorts of nasty formatting issues (paths are missing, formatting is stripped, lists have no bullets). - When committing or creating a PR fails, I get a temp modal with no easy way to get back to the thread that spawned it.

It’s the little things that matter, what are some small papercuts you have noticed in Codex? We’ll fix as many as possible in the next week.

Finally got my swag from the Super Bowl Easter egg giveaway!

@flexibits I've always liked the sound Fantastical (macOS) makes when adding a new event, but it's always WAY too loud relative to other system sounds. It's been this way for years. Can you tone it down?