Why does agent retirement criteria fail?

The most common reason agent retirement criteria fails: Treating retirement criteria as theoretical rather than operational. You read this lesson, nod, and think 'I should probably retire some agents.' But you do not write specific, measurable criteria in advance. Without pre-committed criteria, every retirement decision becomes a real-time judgment call — and real-time judgment calls about your own systems are exactly where sunk cost bias, identity attachment, and status quo preference are strongest. The criteria must exist before the retirement question arises, not after.

The fix: Select three cognitive agents you currently run — habits, routines, decision protocols, or review systems. For each one, write three retirement criteria: one based on performance (a measurable decline in the output that originally justified the agent), one based on relevance (a change in your context that makes the agent's purpose obsolete), and one based on cost (the maintenance burden exceeding the value produced). Set a calendar reminder for 30 days from now to evaluate each agent against its criteria. If any criterion is met, formally decide: retire, replace, or deliberately recommit with a written justification.

The underlying principle is straightforward: Define clear criteria for when an agent should be retired rather than maintained. Without explicit retirement criteria set in advance, you will hold onto agents long past the point where they serve you — because the sunk cost of building them, the identity you attached to them, and the absence of a forcing function all conspire to keep dead agents on life support.