AI Writes Code Faster Than Anyone Can Review — Now What?

AI assistants are helping software engineers write code at incredible speed. In the past, every line of code was checked by a human before it went live. Now, a new study finds that just doesn't work anymore. The authors interviewed professional developers and found that teams are changing how they supervise AI-generated code. Instead of relying only on human review, they layer different types of checks on top of each other.

The first layer is preventive: teams turn their coding rules and architectural goals into machine-readable formats, so the AI follows them automatically. The second layer is executable: automated tests and quality pipelines (checks that run code to find bugs) now catch most errors without needing a person. The third layer is human oversight: reviewers stop reading every line and instead look at the whole design, asking questions like 'Is this future-proof?' and 'Does this fit our long-term plans?' Confusingly, this means fewer human eyes on each line, but more focus on the overall picture.

For everyday users, this could mean faster app updates, more features, and lower software costs — because companies can ship code quicker without sacrificing quality. But there's a tradeoff. If automated tests miss something, fewer people are around to catch it. That puts more pressure on teams to write really good tests and on leaders to trust their automated systems. It also changes the job of a software engineer: less grunt work, more architecture and supervision.

This study is small — just five practitioners — but it points to a clear trend. As AI generates more code, human review can't scale. The way forward is layered supervision: machines check the details, while humans manage the bigger picture. It's not AI replacing developers; it's everyone changing roles to keep up with the pace.