IEI Article
Pricing Machine Decisions: Toward Markets for Autonomous Work
If agents are to allocate scarce resources efficiently, they need price signals built for machine-speed decisions. We examine what a market for autonomous work would require — and where today's mechanisms fall short.
- Markets
- Mechanism Design
- Pricing
Markets are, among other things, a compression algorithm: they collapse an enormous amount of distributed information into a single number — a price — that participants can act on without understanding everything behind it. For human markets, the compression runs at human speed. For an economy of autonomous agents, the cadence, granularity, and trust assumptions are different, and the existing mechanisms strain.
Why the cloud pricing model doesn't transfer
The reflexive answer is that we already price machine work: cloud compute is metered and billed by the second. But cloud pricing is administered, not discovered. A provider sets a rate card; consumers take it or leave it. That is adequate when supply is abundant and substitutable, and inadequate the moment agents compete for genuinely scarce, differentiated resources — a specific model's attention, a rate-limited data source, a human-in-the-loop approval.
For those, you want a price that clears, not a price that is posted.
Three frictions specific to agentic markets
- Latency of commitment. A human can deliberate for minutes; an agent may need to commit in milliseconds. Auction designs that assume patient bidders break down.
- Verifiability of delivery. When the good is "a decision" or "a unit of reasoning," how does the buyer verify it was delivered as specified? Markets presuppose that quality is observable; here it often isn't.
- Principal alignment. An agent bidding in a market is spending someone else's money under delegated authority. The mechanism has to respect budget constraints and mandates set by a principal who isn't in the loop.
A research agenda, not a product
We are not proposing a marketplace. We are proposing that the mechanism-design questions be taken seriously before the marketplaces arrive — because they are arriving, and the default will otherwise be administered pricing dressed up as a market.
The Institute's pricing work sits deliberately at the intersection of mechanism design, financial market microstructure, and AI systems. None of those fields alone has the full picture. Convening them is part of the point.