We started using AI coding tools about a year ago, and the output is undeniable—code gets written faster. But I've noticed something strange: our delivery speed hasn't improved proportionally. After some reflection, I think I know why.
Code review has become the new bottleneck. Senior engineers on my team are still spending just as much time reviewing agent-generated code as they did reviewing human-written code. Maybe more, actually, because the code tends to be longer and sometimes makes unusual choices. The quality is fine, but reviewing it thoroughly still takes time.
This got me thinking about what to build with AI in this space. The obvious opportunities are better automated review tools that can handle what humans are currently doing manually—consistency checks, security scanning, style enforcement at scale. But I'm also curious about process-level changes: how teams actually adapt their workflows when review becomes the constraint instead of writing.
Has anyone successfully shipped code without human review in an AI-assisted workflow? I'm skeptical but genuinely curious if that's working somewhere.
Top comments (0)