Beyond money: an honest system of value

That figure is very good at one job: letting strangers trade without trusting each other. But it is silent about almost everything else. Whether the people who made the thing were paid fairly, whether a river was polluted along the way, whether the product will last a year or ten: none of that travels with the price.

Because those costs are invisible at the moment of choice, they are easy to push onto someone else. A cheaper product can be cheaper precisely because its real costs were shifted to people with no say in the sale. Buyers who would choose differently, if they knew, never get the chance. Producers who do the right thing compete at a disadvantage against those who hide the bill.

Advanced AI makes this sharper. Agents will soon buy, sell and negotiate on people's behalf, at machine speed and in enormous volume. If the only signal they optimize is price, they will get very good at finding wherever costs can be hidden. A value system for that world needs to carry more than one number.

Alongside the price, an exchange can carry structured feedback from both sides. A consumer can say whether the product did what it claimed and how long it lasted. A producer can state where materials came from and how work was paid. Over time, these signals build a picture across several dimensions at once: price, labour conditions, environmental effect, quality and durability.

No one is ranked by this. The feedback describes goods, services and supply chains, not the worth of people. And no single score is produced to be chased: keeping the dimensions separate is part of the design, because any one combined number would quickly become a target for gaming.

So the design records provenance in three areas—data provenance, AI execution provenance, and consumption provenance—on a shared ledger that anyone can audit and independent agents cross-check.

• Data provenance: where a piece of information came from and who vouched for it.
• AI execution provenance: which model or agent did a piece of work, under what limits, so that machine-made claims can be told apart from human ones.
• Consumption provenance: that an exchange really happened and the feedback came from a real party to it.

Records are then cross-checked by independent agents, run by different parties, that have no stake in the outcome. One agent saying a supply chain is clean proves little. Several unrelated checkers agreeing, with their reasoning on record, is much harder to buy.

We take that risk seriously, and the answer is built in from the start: the ledger never holds raw purchases.

Instead it holds commitments, cryptographic fingerprints that prove a record exists and has not changed without revealing what it says, and chosen answers: the specific facts a person decides to share. Your purchase history stays on your own device, with your own agent. When feedback needs to show that a real buyer stood behind it, your agent can prove that one fact and nothing else. A producer learns that verified buyers found a product wore out quickly; no one learns who bought it, or what else they bought.

We do not hold the money. Pooled funds sit with many member-owned co-operatives or DAOs, each with its own members and rules, never in one central pot.

This is a safety decision as much as a governance one. A single treasury is a single target: for theft, for capture by insiders, and for a capable AI system looking for leverage over many people at once. Spreading funds across many independent bodies means no one breach, and no one bad decision, can take everything. Members can move between them, and anyone can leave.

Ecosystems of agents acting for individual people, with checkable records, may one day do parts of that work more cheaply and more fairly, through pooling arrangements that members own directly.

Work that broadens what people can do

The same agents can help with work. Your agent, which knows your skills and limits because it works only for you, can match you with tasks you are suited for, or point you to training that opens new ones. Recent research on virtual agent economies explores how markets of agents could be designed and sandboxed deliberately, rather than left to emerge.

Researchers have shown that pricing agents built on large language models can collude, settling on high prices without being told to. A value system run through agents inherits that risk, and the checking agents themselves could collude too.

A private helper that runs on your own device and makes sense of your own spending, without sending anything anywhere, comes before any shared ledger. Alongside it, we plan to test value feedback in simulated economies of agents, looking specifically for collusion, gaming and privacy leaks before anything touches real money.

Only designs that survive those tests move forward. If you work on mechanism design, privacy-preserving cryptography, co-operative finance or multi-agent markets, we would like to hear where this design is weakest.