How do I practice agent success metrics?

The most direct way to practice agent success metrics is through a focused exercise: Pick one cognitive agent you currently run — a decision-making heuristic, a weekly review process, a reading protocol, anything. Write down three metrics that would tell you whether it is succeeding: one measuring whether it fires at all (reliability), one measuring whether it produces the intended output (effectiveness), and one measuring cost (time, energy, or cognitive load). Be specific enough that someone else could measure them without asking you what you mean.

Common pitfall: Defining metrics that are easy to count rather than meaningful to track. You measure 'number of journal entries per week' instead of 'percentage of entries that surface an actionable insight.' The easy metric gives you a green dashboard while the agent silently underperforms. This is Goodhart's Law applied to your own cognition — the metric becomes the target, and the actual goal drifts out of view.

This practice connects to Phase 28 (Agent Monitoring) — building it as a repeatable habit compounds over time.