Give your agent a stopwatch

Imagine you're a running coach and you have a star athlete on your team.

Now imagine that it’s the first day of training and you both show up to the track. You tell the athlete “run faster.”

What should happen?

• The athlete might say: “faster than what?”
• Or the athlete might run a lap around the track and say: “was that faster?”

And so forth.

You have a fundamental problem: you are the athlete’s only source of feedback and you do not know how to keep time.

This is a dumb way to train, of course.

If you really wanted the athlete to “run faster” the first thing you’d do is get a stopwatch in order to establish a baseline and measure progress.

You are the coach. The athlete is your coding agent.

Today, when you tell an agent to "make this faster" or "fix the bugs," it will try things and asks you to check its work. You are the feedback loop. But guess what: you're slow, expensive, and unreliable. You context-switch, miss things, get tired.

Every time an agent asks you to check something an alarm should go off in your mind—the agent is asking you to be the feedback loop. You should pause and ask: what tool is the agent missing that would let it self-verify?

This is the mental model shift: agents are only as good as their feedback loops. If you want them to work longer and produce better results, you have to give them tools to measure their own work—benchmarks, screenshots, testing frameworks, debug tools, PRDs, todo lists, etc.

Give your agent a stopwatch.