We Know AI Can Run the Mile. How Fast Can It Get?

If I tell you someone can run a mile in ten minutes, you know they can run a mile. What you don’t know is the thing that actually matters: can they run it in six? In five? What is their capacity to improve?

We Are Asking AI Agents the Wrong Question

Organizations are racing to build and buy AI agents. Most of the evaluation stops at the first question: does it work? Does it answer the call, process the document, route the request? That is the ten-minute mile. It tells you the agent can finish the course. It tells you nothing about whether it is performing anywhere near what it could.

That is a familiar problem to anyone trained in Lean Six Sigma. A process that works is the starting point, not the finish line. The discipline exists to answer the next question: how well does it perform, how much does it vary, and how much better can it get?

What a Black Belt Would Measure

The tools already exist. They just haven’t been pointed at AI agents yet.

• A baselineCycle time, first-pass accuracy, escalation rate, rework, and cost per transaction, measured before and after the agent arrives.
• VariationNot just the average, but how widely the agent’s results swing from one case to the next.
• CapabilityWhether the agent consistently performs inside the limits the business actually needs, the way a process capability study does for a production line.
• Control over timeA control chart that catches drift as data, prompts, and models change underneath it.
• HeadroomThe gap between today’s performance and the best demonstrated performance. That gap is the capacity to improve.

A Performance Audit, Not a Compliance Audit

When people hear “audit,” they think of compliance: was a rule followed? This is different. A performance audit asks whether the work is being done at its best possible level, and how to get it there. It is the question Lean has always asked of people and processes. As more work moves from people to agents, someone needs to ask it of the agents.

Agents should be built to check themselves, and the best ones will. But a system checking its own work is still working inside the goal it was given. Independent measurement is what tells you whether that goal, and the agent’s performance against it, is any good.

Measure the Decisions, Too

An agent will do what it is told. If an executive tells it to grow volume at any cost, it will. A Black Belt’s job has never been just to speed up the process; it is to say when the process is aimed at the wrong target. The same independence belongs in an AI performance review: measure the agent, and flag when the direction it was given is working against the business.

Leverage the Builders, Own the Measurement

I don’t think the opportunity is in building yet another agent. The largest technology companies are already doing that, and doing it well. The opportunity is in the discipline that sits on top: Lean Six Sigma methods, applied by agents and people together, that pressure-test AI performance across an organization and keep raising the bar.

We know AI can run the mile. The next few years will be about finding out how fast it can get, and who is measuring.