If AI Writes the Code, Why Are We Paying to Review It?

A recent TechCrunch article caught my eye: 👉 https://techcrunch.com/2026/03/09/anthropic-launches-code-review-tool-to-check-flood-of-ai-generated-code/

It covers Anthropic’s AI-powered pull request review tool, priced around $15–$25 per review. At first glance, that seems reasonable—until you think about how this actually gets used.

A lot of this isn’t reviewing human-written code. It’s reviewing AI-generated code. So the workflow starts to look like: generate code for a few dollars, review it for ~$20, fix what’s found, then pay another ~$20 to re-review. Now a single iteration loop is pushing $40–$60+ per PR… for issues we expect to be there.

I’ve seen this kind of thing before. Years ago, we had a sorting bug that passed review because it looked right, but broke under a specific dataset. Exactly the kind of edge case AI should catch instantly. But in this model, I’m paying to generate the bug, detect it, fix it, and validate it. That feels off.

So here’s the real question: if AI review finds issues in AI-generated code, should I be paying for that—or is there a world where the product shares accountability (credits, refunds, etc.)? Most won’t go there.

Which makes me wonder if we’re solving the wrong problem. If AI is writing the code, shouldn’t validation happen before the PR even exists? Generate, test, validate, then surface. At that point, reviews are about architecture and intent—not syntax and edge cases.

Feels like we’re still taking a human workflow and inserting AI into each step… instead of redesigning it.

#AI #SoftwareEngineering #CodeReview #DevTools #TechLeadership