Commit

How much code do you manually review? Addy Osmani from Anthropic made a very interesting post on X, showing how the Claude Code team uses AI as the first reviewer of practically everything. This is something many have already noticed. It looks at PRs, searches for bugs, validates whether the change actually solves the problem, checks the impact on the rest of the system, and even performs security reviews. In many cases, when the PR reaches someone, the most common problems have already been found or fixed. Their idea is that engineers shouldn't spend time hunting for silly errors line by line. AI does this very well and faster. So human review stops being "looking for bugs" and becomes "does this make sense?", "is this the best solution?", and "is the risk of this change acceptable?". Humans are still necessary when judgment, context, and responsibility come into play. Things like architectural decisions, important product features, and critical areas like authentication, billing, and infrastructure still need human evaluation. In summary: AI reviews the code, but humans continue to decide and are responsible for the changes. https://x.com/addyosmani/status/2114604912520921136…