How do you make an agent relentless instead of quitting at the first obstacle?
A builder complains that agents quit at the first obstacle and hand the work back. The fix is not asking the agent to try harder but giving it a loop it cannot exit early: a clear done-condition it can check itself, permission to try a different approach when one fails, and a rule that reporting what was tried and what blocked it beats saying I cannot. Agents quit when the goal is vague; a checkable finish line keeps them going.