Glossary
Agent Cost of Failure
Agent Cost of Failure is what failed attempts, retries, dead ends and unnecessary repetition added to the cost of an outcome an AI agent reports as successful.
A successful result says the job finished. It does not say what it took to get there. Agent Cost of Failure is the managerial question behind that gap: success and efficiency are different questions, and a passing outcome can hide a wasteful route.
Example
Consider an agent that reports a successful deployment: the service is running and all checks passed. Buried in its summary is one sentence: the first virtual machine had failed, so the agent deleted it and built another. Rebuilding may have been the safest route. But the model responses, commands, downloads, restarts and the rebuild all sit outside the green checkmark. A cloud bill may show compute, an AI provider may show tokens and a ticket may show “resolved”; none of them necessarily shows the combined cost of that one decision.
Measuring it
The number must be measured, not invented. That means connecting mission, agent, sub-agent, execution attempt, tool action, outcome, elapsed time and attributable resource usage, and distinguishing necessary recovery from avoidable rework. In many environments the measurement will at first be incomplete, and that missing coverage should be visible too.
The cost is not always financial. A rebuild can delay another deployment, consume a scarce maintenance window or cross an authorization boundary.