The Agent Economy Has No Auditor

By the Founder of Truence

In 2008 I was an underwriter at Citi. When the market collapsed, I took calls from families who were losing their homes. They were not case numbers. They were people who had trusted numbers that turned out to be fiction. Appraisals, ratings, risk models, an entire economy that had stopped checking its own math. I spent the years after that inside some of the largest banks in the country, helping them rebuild risk management and close the regulatory findings that crisis left behind.

Twenty years in banking taught me one lesson that has never left me. Every financial disaster begins the same way. Someone stops verifying, and everyone else keeps trusting.

I'm watching it happen again. This time it isn't mortgages. It's commerce, and the machines we're handing it to.

AI agents are beginning to run real businesses. They shop for consumers. They sell for merchants. They buy advertising, set budgets, negotiate, and move money. Every week, more decisions that used to pass through human hands pass through an agent instead. The people building this future say commerce is moving beyond conversion and attribution into the age of agents. They're right. It is.

This is not a forecast. Visa and Mastercard have both launched programs that let AI agents make payments. OpenAI and Stripe built checkout directly into ChatGPT. The rails for machines that spend are being laid right now, by the largest payment companies on earth. The rails for checking what those machines did do not exist.

So let me ask the questions nobody is asking.

Do you know what your agent did yesterday? Not what it reported. What it actually did, and what actually settled in your bank account because of it?

What happens when eight out of ten agents are simply bad at their jobs? Untrained, poorly tuned, optimizing the wrong number with perfect confidence? A bad employee loses money until a manager notices. A bad agent loses money at machine speed, around the clock, on every channel at once, and files a beautiful report about it.

What happens when fraud learns to speak agent? Not fraud that steals one card. Fraud that quietly steers the systems doing the buying and selling. How long until a business finds out? How much capital is gone by then?

And here is the question I care most about, because I spent my career inside risk. When banks are asked to finance businesses that are run in part by agents, what proof will they accept? A dashboard? A screenshot? A performance report written by the same agent that spent the money? No underwriter on earth should accept that. Eventually none will.

Commerce already runs on self-reported numbers. The ad platforms grade their own homework. The dashboards repeat what the platforms claim. The books get reconciled months later, if at all. That is the black box merchants have lived in for years, and it costs them margin they never see leaving. Agent commerce doesn't fix the black box. It seals it. The spending, the reporting, and the optimizing all happen inside the machine, and the owner now stands two layers from the truth instead of one.

There is exactly one record in this entire system that cannot be argued with. The bank. Not what a platform attributed. Not what an agent claims. What settled. Money that arrived is the only performance review that matters, and it is the one number almost nothing in commerce is checked against.

In banking we never let anything important run unverified. Every transaction is reconciled. Every model is validated. Every institution is audited. Not because bankers are virtuous, but because we learned what it costs when trust runs ahead of verification. Commerce is now building its most autonomous era on numbers nobody checks, and adding machines that act on those numbers without sleeping.

The agent economy has no auditor. Before it manages much more of the world's money, it needs one. A verification layer where every agent's action, every channel's claim, and every marketing dollar resolves against the bank. So owners know what their machines are doing. So lenders can trust the businesses those machines run.

That conviction is why I built Truence. But whether it's my company or someone else's that becomes that layer, the layer must exist. I took the calls in 2008 from people who paid the price when it didn't.

Verify first. Then automate.