Developers coming from other tools are often impressed by what Codex finds in code review. In this video, Maja Trebacz and I show how to set it up and walk through issues Codex finds in real PRs. Included with ChatGPT Plus/Pro (or ~$1/run with credits).

Setup takes a minute: connect GitHub → enable code review. Codex can review every PR automatically or when you ask for one. Codex Security is also free for the next month! You can also run reviews locally in the Codex app or Codex CLI by typing /review.

@majatrebacz what an amazing feature! glad it’s not $25 👀

@jxnlco @majatrebacz nice. just shipped this using exclusively gpt-5.4. let's see how it reviews its own creation.

@majatrebacz HI, My life doesn't need coding，people like me still need 4o back !#keep4o #bringback4o

@OpenAI @majatrebacz froge cool @FrogeVIP

@majatrebacz $1/run on every PR adds up quick though. what's it catching that a good static analyzer misses?

@majatrebacz #nomasssurveillanceissoobvious #keep4o #keep4oforever #keepyourpromises #FireSamAltman

@OpenAI @majatrebacz @ChatGPTapp give me my money back we should be able to request refunds for degraded service, which is definitely what happened after 4o

@majatrebacz /review in codex is game changer.

@majatrebacz Thank you~♡ OpenAI Korea B2C "Me"

@majatrebacz I'd watch – execution on these code reviews could seriously impact dev velocity.

@OpenAI @majatrebacz Instead of focusing on codex,you should keep it in your heart

@majatrebacz Seeing AI highlight real problems in PRs is often eye-opening for developers coming from traditional workflows.

@majatrebacz Wow.

@majatrebacz automatic review gets useful the day it catches the bug you were proud of

@OpenAI @majatrebacz So not $15-$20/pr like Claude? 👀

@majatrebacz Codex its an amazing tool. I use it everyday. Thank you for keep making it better everyweek!!

@jxnlco @majatrebacz intent matching is the killer feature. catching when a diff doesn't do what the PR says it does

@majatrebacz scammer openai scammer $wld scammer sam altman

@majatrebacz Been watching more devs discover these code review tools - the pattern recognition stuff is getting pretty solid

@majatrebacz I'm impressed about how many fucking times I have to correct that piece of shit software. Once my subscription is over I'm jumping ship, it's not worth the low price to have weeks of work gone in one fucking prompt

@majatrebacz code review is a great entry point for AI agents because the feedback loop is fast and the blast radius is low — you can improve signal quality over time without risking prod issues. interesting to see this pattern getting more practical and accessible.

@majatrebacz solid!

@OpenAI @majatrebacz Thanks, Romain! Hope it’s worth stuffing your pockets while working for a complete and utter ghoul (I’d imagine if you’re still in Sam Altman employment you may, likewise, be a ghoul)

It’s interesting to see Alex leaning into Codex over Claude Code lately. With ChatGPT 5.4’s logic and those higher rate limits, it’s becoming a beast for full-app generation. The hybrid workflow is the real future - using Claude’s /btw for quick side logic while letting Codex handle the heavy lifting of the repo. We’re getting to a point where the “engine” you choose is like picking a favorite car brand, they’re all fast - it’s just about the feel.

@majatrebacz Genuinely curious: with PT Plus/Pro (or ~/run with credits)., what are you optimizing for?

@majatrebacz scam worldcoin thief sam altman...

@majatrebacz

@majatrebacz This is huge for DeFi development. Smart contract audits cost $50K+ and take weeks. AI-powered code review as a first pass could dramatically improve security and speed for treasury protocols.

@majatrebacz is the model in cloud the same with the one in /review on cli? the credits for cloud seems different track

@OpenAIDevs @majatrebacz This thing doesn’t get tired and actually reads intent. Game changer. Thanks 💪

@majatrebacz This is kind of big if the 'review' power surpassed the 'bug' power AI coding agent produce, then this becomes a complete circle and the need for human to remain in the loop is greatly reduced.