They ignore your AGENTS.md and colour your checkout buttons with raw hex values, reach deep down into your database connection to delete the data which they can't handle, and write a 1000-test suite that succeeds no matter what the code does.
But whose name is on the PR?
Yours.
You've probably already tried spec-driven development, but Markdown files are just more guidelines for agents to ignore; having other agents review the agent-written code, but that's a never-ending story — and still without guarantees; CI checks... but you've blown through your subscription token limits by the time they fire.
So you watch the agents like a hawk, because you can't bring yourself to "LGTM" what you are accountable for but haven't reviewed.
No wonder you're exhausted.
What you actually need is an encoding and verification of your rules outside the agents, so you have full control, and not just the accountability; deterministic checks, so you can trust them; and to have them right in the generation, so the agents do it correctly the first time.
We believe coding with AI should leave you feeling calm, competent, and in flow, instead of frustrated, scattered, and drained.