Tal Weltsch retweetledi

I think Anthropic identified the right problem. Code review is where everything slows down now. AI made writing code fast, so the bottleneck moved to reviewing it.
But bug hunting is probably the easy part of review. And that's mostly what they’re showing here..
The bugs in AI-generated code are usually obvious. The hard questions are different. Does this follow the patterns your team agreed on? Does it make sense for where the product is heading next month? That context lives in developers' heads, not in any file the AI can read.
And there's a trust paradox here. Production ready basically means two things: someone you trust reviewed it, and it contains solid tests. AI only replaces the review step when the person you trust is the AI.
If you trust Claude enough to write your code, why do you need a separate Claude to review it? If you don't, putting the review in the cloud doesn't change anything, it just makes it feel more “trustworthy" and this is what I read from this move from anthropic, a play for trust.
By the way, one place where none of this applies: product experiments. Those should be cheap and fast, no hardening needed. That's where AI shines without the review problem and this is one of the reasons we build @dazl_dev.
There's always a bottleneck somewhere.
P.S why are we catching bugs at review time? If the AI is writing the code, bugs should be caught before the PR, not after. That's a weird time to solve this issue.
Claude@claudeai
Introducing Code Review, a new feature for Claude Code. When a PR opens, Claude dispatches a team of agents to hunt for bugs.
English



















