The review problem is your problem too

Anthropic shipped a new feature yesterday called Code Review. It sends a team of AI agents to check every pull request — the mechanism developers use to submit code changes before they go live — for bugs that human reviewers miss. It costs $15–25 per review. It's aimed at enterprise engineering teams at companies like Uber and Salesforce.

That's the headline. Here's what it actually means.

Their own engineers are producing 200% more code than a year ago, thanks to Claude Code. The code is coming so fast that the people responsible for checking it can't keep up. Before this tool, only 16% of code changes got serious feedback from a human reviewer. The rest shipped with a skim and a prayer.

So Anthropic built AI to review the code that AI wrote. Read that sentence again.

This is the part that matters if you're not a developer. The bottleneck in software has moved. It's not writing code anymore. It hasn't been for a while, but now even the people who build these tools are saying it out loud. The constraint is knowing whether the code does what you intended. Judgment. Context. Understanding the problem well enough to know if the solution is right.

That's not a programming skill. That's a thinking skill. Product thinking. Domain expertise. The thing where you look at what came back and say "that's not what I meant" or "that would break the onboarding flow" or "users will hate this on mobile." The stuff that lives in your head because you've spent fifteen years in your field.

The $15–25 per review price tag tells you something too. Anthropic isn't selling speed. They're selling depth. The reviews take twenty minutes on average — deliberately slow — because thoroughness is what's scarce now, not velocity. Writing code is cheap. Knowing if it's correct is expensive.

For the builders reading this — the PMs, the domain experts, the people shipping with AI tools and no CS degree — this is confirmation of the bet you already made. You're not at a disadvantage because you can't write code from scratch. You're building the same way everyone else is. The difference is vanishing. The advantage is shifting toward the people who know what should get built and can tell when it's wrong.

The engineers at Anthropic need AI to check AI's work. You need the same thing, just at a different scale. The skill isn't coding. The skill is knowing what right looks like.

That's the one thing the tools can't replace.