Matthew Boston

Coding agents need a faster feedback loop

March 6, 2026

Key takeaway: Make linting and the relevant tests run locally in seconds before you hand real work to a coding agent.

Coding agents need fast feedback even more than people do. If a test run takes minutes, the agent can’t safely explore or check its work.

A person can check email while tests run. An agent sits there burning tokens until the result comes back. A five-minute suite is five minutes of dead time per iteration, and a complex feature has dozens of iterations.

The bigger problem is what happens in the meantime. Agents make mistakes. That’s fine if they catch them fast. With a fast loop, the agent sees the broken test in seconds and fixes it. With a slow one, it has already written three more files on top of a broken change.

Aim for these:

  • Linting under 5 seconds for the changed files
  • Formatting under 2 seconds
  • Unit tests under 30 seconds for the relevant files
  • Incremental type checks under 10 seconds

A 10-minute full suite is fine for CI. The agent just needs a way to run the relevant tests in seconds, with test file matching, watch mode, or incremental compilation. And keep network requests out of the core suite. A flaky test is the worst kind of slow feedback.

Every minute you spend speeding up the local loop helps every agent session after it. Agents can afford to experiment, and they catch problems earlier. Speed is only part of it, though. Which checks go in the loop, and in what order, matters too. I covered that in Build the harness before you hand the agent real work.