[latand6]

A lot of the exhaustion comes from agents that improvise on every task. You give it a vague prompt, it takes a novel approach, you review it, half of it is wrong, you correct it, repeat.

I started using and writing skills for the tasks extensively. PR review, deployment, frontend design, that kind of thing. The agent follows a tested process instead of guessing and leads to fewer surprises.

The tiredness is real. But running every agent interaction as a fresh conversation when you could reuse a workflow makes it worse than it needs to be.