A goal brief for an AI agent should define the finish line

An instruction like “improve onboarding” gives an agent a direction, but leaves important choices open. Which users? Which part of onboarding? What evidence should be checked? What should happen if th