OK Codex is GOAT at finding bugs and finding plan errors

@garrytan i think currently openai is providing best value for money in a limited $20… unlike anthropic’s $20

@garrytan The killer workflow right now: Claude for implementation, Codex for review. Two models catching each other's blind spots. Better than any human code review process I've seen.

@garrytan I mean, you can use Claude too for finding Codex's blind spots. It goes the other way around.

Pick which twin will win! You could score a share of $2 million in Bonus Bets.

@garrytan @paulg Well, I ported gstack to be used on codex. Thanks for it🙌🏾 - github.com/devwithjp/gsta…

@garrytan @paulg It’s so good but it’s also ridiculously painful to get a clean review. Get ready to say fix it 20 times. Claude just can’t deliver an implementation that passes codex review. I think this a big area for open Ai to solve. 80% of our token usage on review/fix cycle.

@garrytan this is the underrated part — plan errors are way more expensive than bugs lol a bug costs you an afternoon. a wrong architectural decision costs you months. @OpenAI Codex catching plan errors is actually the bigger story here

@garrytan Not a single doubt codex is doing really good I hae seen claude code has been underperforming at various stages

@garrytan @paulg Will look into that but opus is the real king

@garrytan @jordanjhamel @Lovable has been preety good at this recently

@garrytan @TheRealAdamG #Amen Welcome to the Church of Codex @garrytan, long live Papa @thsottiaux

@garrytan Finding bugs is the easy part. Knowing which bugs matter in your system is the hard part. Codex spots a null pointer. But can it tell you that null only happens when sales uses the import tool on Thursdays because of a workflow nobody documented?

Free Cone Day is here! Come celebrate with us at a DQ location on today by getting a FREE small vanilla cone 🍦

@garrytan Even codex 5.2 was very good at bug fixes and code reviews! $20 plan is so much worth it just for this purpose.

@garrytan @grok generate a bug-hunter movie poster where Codex is a forensic detective circling root-cause clues on a whiteboard at 3am.

@garrytan So finally Open AI shipped something the Tech community can be proud of. All this while the arena was fully owned by Claude.

@garrytan sounds like you're having a relationship with AI at this point

@garrytan Curious if you've tried Codex for other tasks and found it more useful than CC?

@garrytan @paulg Codex is the goat. I don’t know what they’re feeding it but it’s definitely working.

@garrytan may be just use codex all the time? works really well as your codebase grows big.

@garrytan Claude + Codex = AI dev dream team

@garrytan voratiq.com/leaderboard/

@garrytan @paulg Absolutely! it catches a plan error written by Claude Opus.

@garrytan Claude for implementation and Codex for review has been working great for me

It’s time to dance! Get it on tournament action with Bonus Bets from FanDuel.

@garrytan the bug-finding wins because it reads the intent of the plan, not just the syntax. claude writes 500 lines that work — codex reads those 500 lines and asks "wait, is this actually what we wanted?" different questions, different value.

@garrytan plan errors are the expensive ones 10 min of Codex review saves 3 hours of "why is this broken" debugging later

@garrytan How are you managing the cross provider access to each of these agents. Any platforms you’re using?

@garrytan It’s so good man

@garrytan @paulg claude for writing, codex for review!

@garrytan and for the other coding tasks, would you prefer Opus 4.6?

@garrytan Love to see this! 🙌 It still surprises me what it catches.

Dance Into Spring: Haiyang Yangko Fest! During the Spring Festival, 20 Yangko dance troupes from across Shandong Province and beyond gathered in Haiyang, Yantai, for an electrifying performance. With vibrant costumes and energetic moves, they brought the streets to life, spreading festive cheer and welcoming the new year. Tap into the rhythm of tradition. (Source: Yantai Information Office)

@garrytan Codex is awesome at everything. Except probably design haha

@garrytan Codex is brutal in the best way: it catches the bug, then points at the plan assumption that created the bug in the first place. That second hit is where velocity comes from.

@garrytan Been using Codex for a week now and honestly it catches stuff I would've spent hours looking for. Bug finding alone makes it worth it.

@garrytan @paulg 🙂‍↕️ it’s the goat in general and way more efficient with token usage

@garrytan Facts Codex is elite at spotting bugs and fixing logic fast.

@garrytan U mean the model or the app?