What does it mean that agent drift?

Agents degrade over time unless actively maintained — monitoring catches drift before it becomes failure.

Example: You built a morning review habit six months ago: wake up, open your task manager, spend 15 minutes prioritizing the day. It worked perfectly for weeks. Then one morning you checked email first — just once. Then you started reviewing while eating breakfast, skimming instead of thinking. Then you began skipping weekends, then Mondays. Today you open the task manager maybe twice a week, skim it for 30 seconds, and close it without changing anything. No single day felt like a failure. Each deviation was minor. But the agent that once drove your daily prioritization has drifted from a 15-minute deliberate practice into a vestigial gesture. The drift happened in plain sight, invisible because it was gradual.

Try this: Pick one cognitive agent — a habit, routine, or decision protocol — that you have been running for at least 30 days. Write down its original specification: what triggers it, what steps it includes, what output it produces, and how long it takes. Then, honestly describe what you actually did the last three times it fired. Compare the two descriptions side by side. Measure the gap: steps skipped, duration shortened, output degraded. If the gap is more than 20%, you have active drift. Write one sentence naming the drift pattern you see.