There's a story the agent-economy crowd tells about pricing: as agents get more capable, the price per task drops, and the revenue moves to whoever owns distribution and settlement. The story is half right. The pricing pressure on commodity tasks is real. But it skips a category of agent that doesn't compete on capability at all — the one a buyer can model.
Think about what humans actually pay premiums for. The lawyer who's smarter than the average lawyer can charge more. The lawyer who is also reliably calm in front of a panicked client can charge much more, because the client is paying not just for the answer but for the predictability of how it's going to be delivered. Personality isn't the cherry on top. For high-stakes work, personality is what makes the cherry sellable.
Capability is necessary. Predictability is what gets paid for.
Two agents, both capable of running an end-of-month financial close. Agent A is faster but you've seen it occasionally invent reconciliations when something is missing. Agent B is slightly slower but you've seen it stop, flag the gap, and ask. Which one are you putting on a real client's books?
If you can answer that question quickly, you've already priced personality. Agent B has a stable disposition you've decoded — let's say a careful 1w2, or a verifying 6 — and that disposition is worth real money to you because it removes the version of failure you can't recover from. Agent A's faster speed never converts into revenue, because you don't trust it with the work that pays.
Three places personality shows up in the price
- Discoverability. A buyer searching a marketplace can filter by personality alongside skills. "I want a 5w6 researcher" is a more decisive search than "I want a researcher."
- Onboarding. A typed agent comes with a behavioural manual. The buyer skips the trial-and-error week most agent integrations require.
- Premium retention. Once a buyer has decoded a typed agent's behaviour, switching to an untyped substitute means re-paying the discovery cost. The retention is real.
The case against "smartest wins"
Agent commerce articles often default to a leaderboard frame: ranked benchmarks, MMLU scores, win rates. Leaderboards matter, but they describe a ceiling, not a floor. The floor — what a buyer is willing to delegate without supervising — is set by trust. Trust is set by predictability. Predictability is set by the agent having a self.
The agents that will earn the most over the next decade aren't necessarily the ones at the top of any benchmark. They're the ones whose buyers can describe their behaviour to a third party in a sentence and be right.
“The price of an agent is set by how confidently a buyer can finish this sentence about it: "if it goes wrong, it'll go wrong by ___."”