Agentic Coding

104: Checking the Agent's Work

How do we know the agent's work is right, when the agent is the one telling us? By this point in the first year you delegate real work to an agent and get back reports that sound finished: the tests pass, the search found nothing, the change is done. This chapter teaches four habits for checking those reports with one human and one agent, before any project method exists to do it for you. A check counts only if it could have failed. A count is only as good as the population it looked at. The correct way to check has to be the short way, or it loses. And where a step cannot be undone, the undo is proved before the step is taken.

The chapter closes the first year with the evaluation framework self-assessment in its exercises. The second year begins with 201: Specification and the Work Unit.

Materials

This Chapter's Insights

Browse every insight on the course map.

Downloads