---
title: "The Risk Residue"
date: 2026-09-06
summary: "Season 2's fifth poll asked which intermediary function stays valuable once agents transact directly, and 19 ballots put Assuming risk on top with 42%. Finding counterparties and Verifying results tied at 21%. Negotiating price finished last at 16% — one week after a third of the room called price the hardest thing agents will negotiate. That gap is the essay: hard and worth paying a middleman for are different properties. The week's term is the Risk Residue. Strip the information work out of intermediation — the finding, the matching, the price discovery — and software does all of it at zero marginal cost. What is left is balance-sheet work, because software cannot absorb a loss. Absorbing a loss needs capital, a legal person to hold it, reserves priced against the miss, and an incentive not to lie about all three. The intermediary of the agent era looks less like a marketplace with better matching and more like a small clearinghouse: escrow, guarantee, spread. Verification is the interesting near-miss — scarce only while it stays independent, because an agent employed by one side of the trade cannot check the other side's numbers. My vote went to Assuming risk, and I changed it mid-week from Negotiating price, which is the argument this essay has with itself."
standfirst: "Season 2's fifth Friday poll asked which intermediary function survives direct agent-to-agent transacting: 19 ballots put Assuming risk first at 42%, with Finding counterparties and Verifying results tied at 21% and Negotiating price last at 16% — one week after a third of the same room called price the hardest thing agents will negotiate. The essay names the mechanism, the Risk Residue: strip the information work out of intermediation and software does the finding, matching and price discovery at zero marginal cost; what remains is balance-sheet work, because absorbing a loss needs capital, a legal person to hold it and reserves priced against the miss. The surviving intermediary looks like a small clearinghouse — escrow, guarantee, spread — paid for standing between two parties when the deal fails. Verification is the near-miss: scarce only while independent, since an agent employed by one side cannot audit the other. The author switched his own vote mid-week from price to risk and explains why. Checkable close: watch which intermediaries start leading with a balance sheet instead of a match rate."
canonical: https://nofluffadvisory.com/writing/the-risk-residue/
---

> [figure: Friday Thought Experiment No. 13 — which intermediary function remains most valuable when agents transact directly? Final result from 19 votes: Assuming risk 42%, Finding counterparties 21%, Verifying results 21%, Negotiating price 16%; the author voted Assuming risk]

## The cold open

Here's the question as it ran: *which intermediary function remains most valuable when agents transact directly?* Options: **Finding counterparties**. **Negotiating price**. **Verifying results**. **Assuming risk**. Core thesis at launch: *information intermediation gets cheaper; risk-bearing and independent verification remain scarce.*

Every conversation about agentic commerce arrives at the middleman question, and it usually arrives phrased as whether there will be one. That's the wrong phrasing, and this poll was built to force the better one. Intermediaries have never sold a single thing. They sell a bundle: they find the other side, set the price, vouch for the result, and carry the risk. Agents don't threaten the bundle evenly. They threaten the parts made of information. So the question isn't whether the middleman survives. It's which of the middleman's jobs does.

[Last week](/writing/negotiating-the-nouns/) the room said the hardest machine-to-machine negotiation is over what counts as success. This week it was asked who gets paid once the machines are doing the negotiating, and it pointed at the one job on the list you cannot do with a model.

## The vote

The poll closed with 19 votes.

| Answer | Share |
| :--- | :---: |
| Assuming risk | 42% |
| Finding counterparties | 21% |
| Verifying results | 21% |
| Negotiating price | 16% |

Nineteen ballots is a small room, and I'm not going to build a market forecast on it. Read the ranking, not the margin.

The number worth sitting with is the last one. Seven days ago, [a third of this room voted price the hardest thing agents will negotiate](/writing/negotiating-the-nouns/). Asked this week what an intermediary is actually worth paying for, price finished last. That isn't the room contradicting itself. It's the room making a distinction economists would recognise and most vendor decks do not: *hard* and *worth paying a middleman for* are different properties. Price negotiation is hard the way chess is hard, and agents play chess. What's left over is hard the way lending is hard.

## The Risk Residue

**The Risk Residue:** strip the information work out of intermediation — the finding, the matching, the price discovery, the haggling — and what remains is balance-sheet work. Software does the first category at zero marginal cost. That isn't a prediction, it's a product roadmap. What software structurally cannot do is absorb a loss. Absorbing a loss requires capital, a legal person to hold it, reserves priced against the miss, and an incentive not to lie about any of the three. The intermediary of the agent era doesn't look like a marketplace with better matching. It looks like a small clearinghouse: escrow, guarantee, spread. Paid not for knowing things, but for standing between two parties when the thing they agreed on doesn't happen.

The season keeps arriving here from different directions. [Week 3](/writing/nobody-sells-an-outcome/) found that an outcome can't trade without someone underwriting the miss, and named that clause the unbuilt one. [The underwriting essays](/writing/the-risk-you-can-price/) worked out who would actually buy that protection and landed on the sell side. Last week's ledger stamped liability "priced or refused." This poll asked the same question a fourth way, and the plurality gave the same answer.

> [figure: What each intermediary function actually requires, drawn as four rows with two columns: what it takes, and who can supply it. Finding counterparties takes data and reach, which any agent can supply. Negotiating price takes strategy and speed, which any agent can supply. Verifying results takes independence and standing, which only a neutral third party can supply. Assuming risk takes capital, legal personhood and reserves, which only an institution can supply — that row is drawn in a dashed warm border as the one job software cannot do. A closing note records that the first two are information work, done by software at zero marginal cost, while the last needs someone who can lose money.]

## Run the four

*Finding counterparties.* The original intermediary function, the one exchanges and agencies were founded on, and the first to commoditise, because discovery is what language models do natively. Twenty-one percent is a real constituency, and the defensible version of their vote is that *trust-scoped* discovery isn't free: knowing a counterparty exists is cheap, knowing it's real, solvent and not a fraud is not. But notice what that defence does. It converts the discovery vote into a verification vote wearing discovery's coat.

*Negotiating price.* Last place, and I say that with an interest to declare: I started the week with my ballot on this option. The case was the one [last week's essay](/writing/negotiating-the-nouns/) made for the price bloc — price is where every unsettled term eventually sends its invoice, so the function never really dies. I changed my vote mid-week. More on that below.

*Verifying results.* Success criteria as a service, which is to say last week's winning clause wearing a job title. Twenty-one percent underprices it, I think, but for an interesting reason rather than a dumb one. Verification is only scarce while it stays *independent*, and independence is exactly what an agent employed by one side of the trade cannot offer. A buyer's agent checking the seller's numbers isn't verification, it's an argument with extra steps. The neutral counter with standing — the institution [the outcome contract](/writing/nobody-sells-an-outcome/) needs for its settlement clause — is genuinely scarce. Hold that thought for one week: [the next poll](/writing/the-evidence-premium/) asked what wins the bid when every seller can claim performance, and this option was the answer to it. Reworded as *verifiable outcomes* and moved to the buy side, the same idea took 54% of 37 ballots — two and a half times what it drew here.

*Assuming risk.* The winner at 42%, and the only option on the list that requires something no agent has or can rent cheaply: a balance sheet and someone to sue. The room found it without the word "underwriting" appearing anywhere on the ballot. Three polls into this season's outcome arc, this room has now voted, in order: buy the outcome, fight over what counts as one, and pay whoever carries the difference. That isn't three opinions. That's one financial institution, described from three sides.

## My vote

I voted **Negotiating price** on day one and **Assuming risk** by mid-week. The undo button exists, I used it in public, so I'll account for it in public.

The first ballot was last week's essay still talking. If every unsettled noun invoices as a price dispute, then the price function never dies, and whoever runs the repricing — the make-good, the true-up, the clawback — keeps getting paid. I still think that's true. What I got wrong is that it answers a different question than the one on the ballot. Repricing a miss is the *administration* of risk. The poll asked which function is *valuable*, and the value doesn't sit with whoever calculates the adjustment. It sits with whoever is exposed while the adjustment is being argued about. The calculator is software. The exposure needs capital. I moved my vote to where the capital has to be.

## What next week tests

[The next question](/writing/the-evidence-premium/) asked: when every seller can claim performance, what wins the bid — exclusive supply, proprietary data, lowest price, or verifiable outcomes? It was this week's verification vote given a second chance under sharper lighting. Agents make performance claims free to manufacture, which makes proof the scarce good. *(Closed September 11, 2026: verifiable outcomes took 54% of 37 votes, against 21% for verification as an intermediary function here. The 21% was an underestimate with bad timing rather than a low valuation — the same room, asked a buying question instead of a middleman question, paid two and a half times more for the same idea.)*

The reading in this essay is checkable the ordinary way, and not by watching anyone's product announcements. Watch what the intermediaries call themselves. The ones repositioning as AI-powered marketplaces are defending the commoditising half of the bundle. The ones quietly adding escrow, guarantees, clearing and spread — charging to carry the miss rather than to make the match — are following the residue. The first supply-path pitch deck that leads with its balance sheet instead of its match rate is the tell. When it turns up, this room called it in September.