About whether AI coding agents can debug running systems.
Has this happened?
Yes 55 (50%)Not sure 34 (31%)No 22 (20%)
Votes cast 1–2 October 2026: 100,590 votes from 9,694 people.
Hacker News has set AI a lot of challenges over the years. Which ones has it met?
About whether AI coding agents can debug running systems.
Votes cast 1–2 October 2026: 100,590 votes from 9,694 people.
Most replies here are about writing code faster. But there's a gap nobody's talking about: AI agents are completely blind to running systems.
When you hit a runtime bug, the agent's only tool is "let me add a print statement and restart". That works for simple cases but it's the exact same log-and-restart loop we fall back to in cloud and containerized environments, just with faster typing.
Where it breaks down: timing-sensitive code, Docker services, anything where restarting changes the conditions you need to reproduce.
I've had debugging sessions where the agent burned through 10+ restart cycles on a bug that would've been obvious if it could just watch the live values.
We've given agents the ability to read and write code. We haven't given them the ability to observe running code. That's a pretty big gap.