Developer Uses AI Agent and Heap Snapshots to Fix 2AM Production Memory Leak

A software developer's Node.js production service crashed overnight due to a slow memory leak that only appeared under real traffic conditions. Initial attempts to fix the issue by asking an AI coding agent to scan source code for suspicious patterns proved ineffective, wasting nearly an hour on incorrect candidates. The breakthrough came when the developer shifted approach — capturing two heap snapshots from the live process twenty minutes apart and feeding the growth diff data directly to the AI agent. With concrete allocation data rather than code guesswork, the agent was able to pinpoint the actual leaking constructor and its source. The experience highlighted that AI coding tools perform significantly better when given real diagnostic evidence rather than being asked to reason from source code alone.