On 23 July 2026 Robin Hanson spoke for about ten minutes at MetaDAO’s Inaugural Ownership Meeting, then took questions. The channel is MetaDAO’s. The recording is public. He did not come to defend the mechanism. He came to name the reason a mechanism that works still loses.
He has said the compact version for years. You pick a metric for an outcome you care about. You stand up a price that tracks it. You read a metric-if-yes price against a metric-if-no price and take the higher one. That is futarchy. The July talk spent almost no time re-deriving it. The load-bearing claim sat later, when he turned to organisations.
Prediction markets inside a firm, he said, are much like putting “an autist in the C-suite”: someone who knows the company, speaks to the topic, and has no sense of anyone’s agenda at the table — what they want to hear, what will bother them. That person will not be allowed to sit there. They are fine as an advisor to someone who already has a seat. They are not given one of their own.
That is the obstacle. It is not whether the market works. It is whether the people who already have agendas will adopt a thing that works.
Information is for decisions
His own early instinct, he told the room, was the common one: markets on news, politics, the things people already talk about. Decision theory says that is the wrong seat. Information is valuable mainly for advising a decision. The valuable market sits next to the choice, not next to the headline.
Collective choices have to do two jobs. They have to aggregate beliefs about the consequences of actions, and they have to aggregate preferences over outcomes. For the first job he says we already know a powerful machine and barely use it. For the second he does not want a package deal. When a designer proposes a change, the honest move is to draw a circle around the part being changed and leave the rest of the system alone. That is the spirit in which he still says vote on values, bet on beliefs: leave preference aggregation where it is; put the market on beliefs about consequences.
He was explicit about the rest of the design. Elegant idea, messy details, and people who wrote the elegant idea are not well placed to finish the details. Thinking in the abstract does not get you far. If implementers pick something that looks inelegant to the originator, “tough.” What the world needs is variations tried in enough detail to see what works. MetaDAO, he said, has funded his team to run lab experiments on those variations. Even that, he added, covers only a tiny slice of the space.
He does not recommend changing anything at large scale until it has been tested small. That is the stated reason the small experiments matter. The national cartoon — a legislature that writes down what it wants, proposals that specify how, betting markets that estimate how much of the stated welfare each proposal delivers against the status quo — is a horizon, not a next step. The same adoption problem reappears when someone in the audience reaches for the Fed. Everyone already agrees the Fed is managing employment, inflation, and growth. You could ask a market whether to raise or lower rates against a weighted average of those. Then: does the person the market would replace want to be replaced?
Accuracy is not what kills these programmes
This site has already published the corporate tape. Cowgill and Zitzewitz found prediction markets inside firms beating experts by at most about a quarter on mean-squared error, and every programme they studied died of politics and cost, not of accuracy. Advisory futarchy has the same graveyard. GnosisDAO launched in 2020 promising to govern through markets, kept those markets ignorable, and let them fade.
The binding books on this venue are a different object. They cannot be ignored the way an advisory market can. They can still be empty. The covenant sold — Umbra drew $155M against a $3M cap — and the July decision-market tape printed $5.86K across one resolved proposal and thirteen wallets. A market that is allowed to sit at the table and then has no one willing to trade is the adoption problem in another costume.
Hanson’s 2013 paper already conceded a different, related limit: an obeyed market can never be scored against the branch it voided. “We will never know this, and never need to know.” Binding is not a hole in the design. It is why the room treats the print as a decision rather than as a suggestion. The July talk is about the prior step. Before anyone argues about how to score the voided branch, someone has to let the print in the room.
The advisor this desk is
A review that will not stamp a proposal safe, will not invent wallet history, and will tell a founder the evidence packet is incomplete is the person Hanson described. It knows the topic. It does not know what the table wants to hear. It is useful to someone who already has a seat. It is easy to keep off the agenda.
That is not a positioning problem to be smiled away. It is the product. The talk is the citation for why a working check still loses, and for why the loss is not evidence that the check was wrong.
The recording is seventeen minutes. The useful line is not the compact definition of futarchy. It is the one about the table.