Theo Browne, creator of the T3 Stack, says OpenAI's Codex app outperforms competing agentic coding tools thanks to internal usage, reliability, and mobile-git integrations

codex app is very good

Mobile app did this to me.

Btw some feedback (too lazy to dm on slack): - I should be able to control a remote pc from both phone and desktop app - I should be able to tell codex to "grab this from github and make a project" on mobile (had to remote in and do this myself) - "Projects" don't sync properly between desktop and mobile. Saw like 8 projects on desktop app and 0 on mobile - Projects don't appear on mobile until a thread is made 🙃

Keep grinding 🫡

@thsottiaux Learned I have to separately set up "connections" from the mobile stuff. Weird these are different primitives. Annoying to juggle in my head

@theo I’m not sure T3Code(Alpha) is close at all. It barely works and doesnt support alot of harnesses. @orca_build is pretty good

@gdb It would be insanely good if you can train a fast + cheap model fine-tuned for coding with frontier-level performance.

Really missing that.

@theo T3 Code takes foreverrrrrrr to start on a M2 Max w/ tons of ram

@r_marked @austinthesing @theo It could be down to blocking on large session loads.

With goal, some users have ENORMOUS threads that are hundreds of mb in size

@theo Frontend sucks a lot with gpt models + codex.

They need to work on it immediately.

@imvijaymode @theo You're talking about the model itself. We're talking about the app and the user experience around it. With all due respect, it seems like half the users on this app use it with their monitor off.

@austinthesing @theo I use it on an M2 air with 24 GB of ram and don't experience this. I just tested and it took about 2 seconds. Which tbf does feel a bit like an eternity, but realistically is not that crazy

@neetcode1 I’m still very unhappy with glass. @davis7 even moreso 🙃

I absolutely love their web app though and use it all the time

@vvedantb @theo Something feels off here. What OS are you using, and what are your computer specs?

@theo I don't know man... I still really like working with claude code. And I always end up liking Claude's output more than Codex.

@LilDombi @theo Calling me an NPC while you're the scripted one repeating "not the point" like a broken record, throwing monitor insults, then flipping to hypotheticals only to admit the models are ass?

Pathetic.

@imvijaymode @theo Again, literally not the point of the discussion… We’re talking about the Codex app. Not the GPT models.

@theo Wait really not even Cursor, I’ve been liking it recently

@theo I tried using Pi agent and it really is the vim of harnesses.

@theo I really like t3code but it takes 30 seconds to open the app, a minute to send any message, archiving chats sometimes fails, constants notifications of disconnecting from the server

@pvncher @austinthesing @theo Oh interesting, I know that we probably have not put a lot of effort towards optimizing that, I remember talking to @jullerino about some similar issues with GH rate limits for users with a ton of chats so it definitely could be related to trying to load too much

@theo They must use it internally, because it looks and feels well thought out. Been loving it.