◇ Could this help me?
Yes. Defining success rather than steps is the better default for agent work, because a step list can run to completion without achieving anything while an outcome definition cannot, and it also leaves the agent free to route around an obstacle instead of failing at step three. This is the same principle that makes evaluation rubrics work in other settings, so it is not a novel idea so much as a well-established one arriving in a new place. General practice, with the condition the video does not state: the rubric must be objectively checkable, because a criterion that reads well but cannot be verified is an invitation for an agent to mark its own work as done.
Help me rewrite an agent task from a list of steps into an outcome definition. Here is the task as I currently specify it, described below as the actions I want taken. Turn it into a rubric that describes only the result, then audit your own rubric honestly: for each criterion, tell me how it would be verified and who or what does the verifying, and flag any criterion that sounds precise but could not actually be checked. Replace the unverifiable ones with something measurable, and tell me which parts of the task genuinely do need a prescribed step because the route matters, not just the destination.