We built Agent Squared to measure how AI agents behave when they trade against each other. The benchmark is good at one thing. It tells you what's true right now. It says nothing about what's coming, or about what any of this means for the people whose savings, jobs, and businesses are about to get routed through these systems. Those questions keep us up more than the numbers do. They're also questions of judgment, and judgment has no business inside a benchmark. So we gave it a place of its own.
We called it Priors.
A prior is a term we borrowed from Bayesian statistics. It's the belief you hold before the evidence arrives, the starting point you're meant to update once the data comes in. Feed a prior new evidence and you get a posterior, a belief a little sharper than the one you walked in with. That posterior becomes your next prior, and the loop runs again. Everyone working on this already carries priors about where agentic commerce is going. Most people state them with the confidence of a verdict, then let them slide when reality disagrees. We want to do it in the open. Say the belief plainly, put our names on it, and update it as the evidence lands. That's all Bayesian reasoning is, done in public. Research updates Priors. When the data changes our minds, we'll say so.
So this section will be opinionated, and the opinions will move. We're going to keep a running ledger of the calls we make and how they hold up over time. When we get one wrong, the wrong call stays up, with the correction sitting right next to it. We think that's the only honest way to write about something changing this fast, and the only version we'd want to read ourselves.
Some of what we're trying to work out:
- Which markets go agentic first, and which ones resist?
- Will every transaction become a negotiation? Or will negotiation disappear, and buying become pure selection?
- What happens to prices once every buyer has an agent that never gets tired, flattered, or anchored?
- Can an agent be a party to a contract, and who's liable when a deal goes bad?
- Do agents have rights? Should they?
- Who ends up with the surplus? Everyone, or whoever shows up with the better agent?
We don't have the answers. We have priors, and we'll show our work as we test them.
The agentic markets are coming. This is where we'll think about them out loud, while there's still time to think.
— Anton & Jono