Why does agent sprawl fail?

The most common reason agent sprawl fails: Adding agents without retiring them, because each new agent passes the individual value test — it produces more value than it costs to run in isolation. The failure is evaluating agents individually rather than evaluating the system. An agent that produces ten minutes of value but adds fifteen minutes of coordination overhead to the rest of the portfolio is a net loss, even though it looks like a net gain when measured alone.

The fix: List every active agent in your current cognitive infrastructure — every habit, routine, system, delegation, or automated process you maintain. For each one, estimate two numbers: (1) the value it produces per week in minutes saved, decisions improved, or outcomes achieved, and (2) the coordination cost per week — time spent maintaining it, feeding it inputs, checking its outputs, resolving conflicts with other agents. Now sum the coordination costs. If the total exceeds 30 percent of the total value produced, you have agent sprawl. Identify the two or three agents with the worst value-to-coordination-cost ratio and ask: would retiring these improve the system even though each one, in isolation, still works?

The underlying principle is straightforward: Too many agents create coordination overhead that can exceed their collective value.