Orchestrating AI code review at scale

Cloudflare just revealed it has been using a small army of AI reviewers to scan code changes before they go through, with up to seven specialist bots checking for security problems, performance issues, missing docs, and rule-breaking. The company says this system has already looked at tens of thousands of internal code changes, can wave through clean work, and can even block releases when it spots something serious. Sounds futuristic, right? The comments section immediately replied: cool story, but are we sure this belongs in the official pipeline at all?

That was the big fight. Several readers argued this kind of AI should stay local and optional, more like a spellcheck before you hit send, not a gatekeeper that can hold up the whole team. One commenter flat-out said non-deterministic AI is basically the opposite of what a reliable automated check should be. Others wanted it moved earlier, like a pre-commit or pre-push tool, to make feedback faster and less noisy. On the flip side, some people were fully sold on the idea and said the return on investment is so huge they’d happily pay for the fanciest model available if it means fewer bugs and less waiting around.

And then came the comedy. One reader joked about the odds of Cloudflare simply buying OpenCode and hopefully keeping it open source. Another chimed in on the painfully relatable problem of AI appearing to “think” forever, turning code review into a suspense thriller where everyone wonders if the job is stuck or just being dramatic. In other words: the product launch was interesting, but the real show was the community asking whether these robot reviewers are heroes, hall monitors, or just very expensive nitpickers.