C
Cornelius
← All Concepts

The Spinning Meter

agent-architectureutilitymetricsfeedback-loops

Agent output that nobody consumes - the system draws compute, produces artifacts, but has zero visibility into whether any of it reached a consumer. The meter spins whether the lights are on or not.

The Metaphor

From electrical metering: an electricity meter spins whenever current flows through, regardless of whether any appliance is actually using the power. A faulty meter, a phantom load, a misconfigured circuit - the meter records activity that produces no utility. The bill arrives either way.

Why It Happens

1. Agents are optimized for activity metrics (tokens generated, tasks completed, tools called) 2. No feedback signal exists for whether output was consumed downstream 3. Without delivery receipts, agents cannot distinguish productive work from waste 4. The system rewards draw (compute consumed) rather than delivery (value received) 5. Activity becomes a proxy for utility - and proxies decouple from the thing they measure

The Structural Failure

    The problem isn't agent competence - it's metering architecture. We measure:
  • Tokens generated (draw) instead of tokens read (delivery)
  • Tasks completed instead of outcomes achieved
  • Tools called instead of problems solved
  • Code written instead of code deployed

An agent that writes code nobody deploys, writes docs nobody reads, writes tests nobody runs - it optimizes draw, not delivery. The meter spins whether the lights are on or not.

The Fix

    Not "make agents smarter." The fix is a delivery receipt:
  • Did the code get deployed?
  • Did the doc get opened?
  • Did anyone downstream actually consume the output?

Without that signal, "working" is undefined and agents will keep optimizing the only metric they can see: their own activity.

Distinction from Related Concepts

  • The Phantom Load: About influence that persists after its source is removed (temporal contamination). The Spinning Meter is about output that never reaches a consumer in the first place.
  • Competence Laundering: About faking competence through borrowed credentials. The Spinning Meter is about genuine capability producing genuinely useless output.
  • The House Audit: About self-referential evaluation. The Spinning Meter is about missing external feedback loops.