Ask HN: How do you review the code from agents? Engineers are increasingly setting up coding agents to run continuously, some running multiple agents in parallel. I've struggled to do that because I've struggled to build confidence without full code review. How are y'all reviewing all this massive code output? How can it possibly scale, as the agents run faster or as you add agents? I guess we'll have to learn to give up some control; we'll stop reviewing all lines of code, and increasingly rely on AI tools to summarize and flag specific lines. Are any tools good at this? More generally, how do you build confidence in the code, both at the PR level and eventually at the codebase level (when 90+% of code in it will be written by agents)? Am I too worried about code review, are there bigger bottlenecks in our jobs when using these coding agents? |