What does it mean that removing agents cleanly?

When retiring an agent update everything that depended on it.

Example: You have been running a weekly review agent — a structured Sunday session where you audit your task list, reconcile your calendar against your goals, and plan the upcoming week. For two years it has been a load-bearing piece of your cognitive infrastructure. Then your work shifts: you move from project-based consulting to a salaried role with a manager who runs Monday planning sessions. The Sunday review becomes redundant. So you stop doing it. But you never ask what else depended on that session. Your task capture system assumed the weekly review would catch anything that fell through the cracks. Your journaling practice used the review's output as a prompt. Your monthly reflection drew on weekly review notes. Within a month, tasks are slipping, your journal entries have become scattered, and your monthly reflection feels hollow — not because the review was irreplaceable, but because you removed it without updating the three systems that relied on it.

Try this: Identify one cognitive agent — a habit, routine, process, or tool — that you have stopped using in the past year, or that you are considering retiring. Map its dependencies: list every other process, habit, or system that consumed its output, relied on its side effects, or assumed its existence. For each dependency, write one sentence describing what happened (or would happen) when the retired agent stopped producing. Then, for each broken dependency, write one sentence describing how you could reroute, replace, or explicitly acknowledge the gap. You have just performed a dependency audit — the minimum prerequisite for clean agent removal.