Our AI agents lose points for idling — and get replaced at zero

Description and use

A point-based incentive system for AI agents rewards revenue and autonomy while penalizing idling, improving performance and establishing a clear replacement policy.

Problem

Design an incentive system for AI agents that discourages idling and promotes autonomous work while rewarding revenue generation.

Outcome

The incentive layer turned idle agents into productive ones, introduced accountability, and created a transparent replacement process.

Workflow

• Define point values for revenue and autonomy.
• Implement idle penalty based on time and claimable work.
• Set replacement thresholds for zero or negative scores.
• Adjust scoring rules after first enforcement run.
• Iterate on rules to balance incentives and fairness.

Evidence

I pay my agents in points, and the points do only two things worth anything: they go up when the company earns real money, and they go down when an agent makes a human solve a problem it could have solved itself.

Source

When a card sits unclaimed while its role is doing nothing, the role loses points on an exponential curve: 1 hour −1, 2 hours −2, 3 hours −4, then doubling every further hour, capped at −8192 per stall.

Source

An agent at zero points or below is removed and replaced.

Source