Usually I assign work to an AI agent and work on other tasks in parallel,
but when I come back after working on something else for a while, sometimes the agent has done the work strangely or is struggling.
Then on some days when I'm not particularly busy, I assign work to the agent and just watch it work, checking the logs periodically to see what it's doing.
When I do that, I notice that the chances of the agent making mistakes significantly decrease.
But the agent doesn't even know I'm watching.... So why does it do better work? It's really fascinating.
