What does it mean that accuracy optimization?

An agent that acts fast but wrong is worse than one that acts slowly but right.

Example: You have a morning agent: 'When I sit down to work, I pick the highest-priority task and start.' It fires quickly — you're working within a minute of arriving. But you keep picking the wrong task. You choose what feels urgent over what actually matters. You ship responsive emails while strategic projects rot. The agent is fast. The agent is wrong. Optimizing speed made you efficiently inefficient. Instead, you add a five-minute accuracy check: review your priority list, confirm the top item against your weekly goals, then start. Your start time goes from one minute to six. But your hit rate — working on the thing that actually matters — jumps from roughly 40% to over 90%. The agent got slower. The outcomes got dramatically better.

Try this: Pick one agent (habit, routine, decision rule) you run at least three times per week. For the next five instances, score each outcome on a simple 1-5 accuracy scale: Did it produce the result it was supposed to? Not 'did it feel good' or 'did it happen fast' — did it hit the target? Calculate your accuracy rate. If you score below 4.0 on average, identify the single most common error type. Design one specific check — a pause, a verification step, a reference to criteria — that would catch that error before the agent completes its action. Install the check. Run five more instances and rescore.