Maximize agent time, minimize your involvement

Many developers are always in the loop with their coding assistant: they prompt an agent, wait 60 seconds, get code back, test it manually, paste errors back into chat, and repeat. You will never achieve high velocity this way.

Frontier developers give agents longer tasks with clear validation steps built in: “implement this feature, write tests, run them, ensure they pass, achieve 90% coverage on new code.” The agent works for 30 minutes or more while you work on something else. Agents will get things wrong along the way, and that’s fine. Let them keep looping, self-correcting, and converging on the right answer.

Keep multiple agents running in parallel on a steady stream of tasks. Review output asynchronously. The constraint on your productivity is no longer typing speed; it’s how many agents you can keep meaningfully busy at once.

As your confidence and tooling grow, push this further: run agents for hours or overnight, have agents kick off other agents, and review results the next morning. The goal is to progressively remove yourself from the loop until your involvement is limited to setting direction and verifying outcomes.