Why does alert thresholds fail?

The most common reason alert thresholds fails: Setting thresholds based on perfectionism rather than reality. If your morning planning agent produces a useful plan 85% of the time and you set your alert threshold at 95%, you'll be in constant investigation mode — treating normal variance as failure. The opposite error is equally dangerous: setting thresholds so loose that only catastrophic failure triggers attention, which means you miss the slow degradation that precedes most real breakdowns. Both errors share the same root cause — setting thresholds without data.

The fix: Pick one agent (a habit, a routine, or a delegation) that you monitor. Write down three numbers: (1) the metric you track (e.g., completion rate, accuracy, time-to-fire), (2) the value you consider 'normal,' and (3) the value that would make you stop and investigate. Now ask: how did I arrive at that third number? If the answer is 'gut feeling,' run the agent for two weeks, record the metric daily, compute the mean and standard deviation, and set your threshold at the mean minus two standard deviations. You now have a threshold grounded in your agent's actual behavior rather than your anxiety about it.

The underlying principle is straightforward: Define clear thresholds that distinguish normal operation from problems requiring your attention.