Jonathan Haas

@Dimillian Sounds about right. Hell of a feature. Thanks for all your hard work on making mobile shine!

/goal implemented goal and it took 45 minutes and 3 turns

@jxnlco thank you loss function, very cool

Thank you gradient descent

@tkkong Gonna be so sad when it’s gone

The 10x Codex limit boost has been the biggest productivity hack

@dkundel Pulling up with a plate 👀 This looks incredible

One of the non-Codex weekend projects 💪 Albacore tuna with fresh nectarines, a mignonette of twice fermented peaches, honey nut squash miso tamari, lime juice and regular tamari. Topped with fried serano chiles, smoked salt, tangerine leaves, chives and micro orchids. 😋

@_Sagiquarius_ @davis7 TIL. Thanks for the tip!

@davis7 Tip: synology file station api works really well. Gives agents much faster and better access to data hosted on the NAS.

Codex desktop app + computer use + GPT-5.5 xHigh is so good Just watched it fix a bunch of container port mapping nonsense on my nas instead of copy pasting screenshots back and forth

@Dimillian gonna give the CUA team a hug. This is awesome.

Same tbh

@0xblacklight we have the technology

can you imagine how much claude would have loved building flash games

@DannyLimanseta @srush_nlp This is beautiful. What a cool thing to make! Thanks for sharing :)

After a couple more days with Composer 2.5, I've got a pretty good sense of what it can do for game dev, specifically this mouse cursor racing game I'm building. It nailed ~80% of the features I threw at it, from planning to execution. Pretty amazing. I did have to switch to Opus 4.7 MAX a few times for features that need stronger visual understanding, like adding a 360 loop to the track or nailing a specific visual effect I had in mind. But man, Opus 4.7 MAX is expensive. Composer 2.5's visual understanding is weaker than Opus 4.7, but for most other things it's pretty damn close, and 10x cheaper. So it's now my default. The beauty of Cursor is I can switch models whenever I want, so I'm never stuck on one. Really excited for the larger incoming model. I have a feeling it's going to shine at planning. Can't wait!

@badlogicgames This is so damn cool. I am so excited for local / on device to take off more, especially as they get more competent. Such exciting times to be alive.

i'm very excited by this.

@thsottiaux Yes please! Especially for goals. Almost all my tasks are goals at this point.

Should we bring batch compute to codex? Aka /slow mode

my commit history this year is 60k commits and my contribution is 'told it to stop being so confident on Tuesdays'

@thsottiaux Thank you Tibo!

Some of you noticed limits drained faster in Codex, we root caused it to an optimization that we rolled back that had an impact on cache hit rates when compacting across long running sessions. We fixed this and have now reset usage limits for all accounts. Enjoy the weekend.

@dianalokada Tokens

>hot dude, nerdy tweet >open profile >codex at openai what are they feeding them there omg

@0xblacklight Thing of beauty. Ty fam

btw github.com/humanlayer/pul…

@yuris Cannot unsee.

This logo is giving mid-2010s vibes

@weswinder Waaaaaay faster.

feels like codex limits got nerfed not sure what caused it this time but they are def burning way faster anybody else seeing this?

@simpsoka 🙏

In all seriousness - we are actively implementing a fix for the context donut, which will be live today.

@Shpigford @hnshah Would love to check this out :) heavy user of markdown doc vaults

looking for 5 more people to give early access to! you'll have a free account for life. gonna start billing next week so last chance to get this guy for free. 🎉

@dunkhippo33 Love this! Building a whole business around this particular type of issue at @EvalOpsDev because of a similar headache (originally from using a very early iMessage + LLM company)

Just open-sourced my first project! AI Validator - a 224-line TypeScript tool that catches AI agents lying about actions they never performed. After running 90+ AI instances in production, we had to solve the "I sent that email" (but didn't) problem. More >>

@StijnSmits I use wallclock time for the same thing. Very useful!

Finally got GPT-5.5 on Codex to think way harder than it normally does. As one would expect it to be 'think very hard', 'ultrathink', 'deep dive' etc - does not work. A simple 'use at least 100 (onehundred) tool calls before coming up with an answer' works really well