Cleaning up after AI rockstar developers

AI coding agents exhibit the same failure modes as "rockstar" developers — optimizing for task completion speed over maintainability, producing code others can't follow, and leaving no institutional knowledge behind. Unlike human rockstars, agents compound the problem by generating tens of thousands of lines per session with zero memory across sessions. Teams adopting agentic coding workflows should treat output as requiring the same scrutiny as inherited legacy code from a departed engineer.