Three Fed Governors Met the Agentic Commerce Industry. The Public Record Is a Guest List.

The Fed's agentic commerce roundtable put Visa, Stripe, Amazon and Cloudflare in one room and published a guest list. Read the roster — and notice who was not invited.

Share
Three Fed Governors Met the Agentic Commerce Industry. The Public Record Is a Guest List.

Convening is a governance model. It is also one you cannot participate in, read the minutes of, or file a comment on.

On 26 May 2026, the Federal Reserve Board held a meeting of its Payments Innovation Roundtable. Three sitting governors attended — Christopher Waller, Michael Barr and Lisa Cook — along with Board staff, the president of the Richmond Fed, and senior staff from Richmond and San Francisco. Across the table sat twenty people from the payments and technology industry.

The topic, per the Board's own record, was the opportunities and challenges of artificial intelligence in payments, with a particular focus on agentic commerce.

That is the entire published account. One sentence of substance, and a list of names.

I have read that list several times now, and I think the list is the document.

Who was in the room

Adyen. Amazon. Amazon Web Services. Centime. Cloudflare. Coinbase. Cross River Bank. Fiserv. Forte Fintech. Google. Lead Bank. Mastercard. Paze. Plaid. Shopify. Skyfire. Stripe. Taktile. Visa.

Read it as a payments person rather than as a journalist and some things come into focus quickly.

Both card networks are there. Both of the largest merchant acquirers and processors in the United States are there, in Fiserv and Stripe, with Adyen representing the enterprise end. Three of the largest technology platforms are there, two of them — Amazon and Shopify — as merchants of record for a substantial share of American online commerce. Coinbase is the stablecoin seat. Plaid is the data-access seat. Paze is the bank-owned wallet consortium.

And then the interesting ones.

Skyfire is a company building payments for AI agents. It was in a room with Visa and Mastercard. The Board deliberately seated the insurgent alongside the incumbents, which tells you the agenda was not a briefing. You do not need both in the room to be briefed.

Cross River Bank and Lead Bank are there, and no other banks are. Not a money-center bank, not a regional, not an association representing community institutions. The two banks in the room are the two that specialise in sponsoring fintechs. That is a meaningful choice about which part of the banking system is treated as the relevant part for this question, and I do not think it was accidental.

Taktile does decisioning infrastructure. Centime does working capital and treasury for mid-market businesses. Both are there, I would guess, because somebody understood that agentic commerce is not only a consumer checkout problem — it is also a question about who approves what, against which limits, with what audit trail. That is a risk-decisioning problem before it is a payments problem.

And then the one that stopped me.

Why is Cloudflare at a payments roundtable

Cloudflare does not issue, acquire, process, or settle. It sits in front of a very large fraction of the public internet and decides which requests are allowed to reach the servers behind it.

Which makes it, for agentic commerce, possibly the most consequential company in the room.

Think about what an agent actually has to do to buy something. Before any payment credential is presented, before any network decides whether to authorise, the agent has to reach the merchant at all. It has to fetch the product page, read the price, find the checkout. And for twenty years the entire web infrastructure industry has been building increasingly sophisticated machinery to stop exactly that: automated clients pretending to be people.

Bot management is not an edge case for agentic commerce. It is the first gate, and it sits upstream of every payments protocol anyone is arguing about. You can agree on a credential format, settle liability, standardise the authorisation message — and none of it matters if the merchant's edge returns a challenge page to the agent.

So the chokepoint for agentic commerce may not be a payment network at all. It may be a bot policy, set by a handful of infrastructure companies, applied by default to millions of sites whose operators never thought about it. That is an enormous amount of power over the shape of a market, held by firms nobody regulates as financial infrastructure because they are not financial infrastructure.

The Board put that company in the room. I think that was perceptive, and I think almost nobody writing about agentic payments has noticed that the problem is partly a web-identity problem.

Who was not in the room

No consumer organisation. No state attorney general. No merchant trade association — Amazon and Shopify are merchants, but they are not the National Retail Federation and their interests are not a corner store's interests. No labour representation, which matters more than it sounds when the subject is systems that transact on people's behalf. No credit union body. No community bank association. And nobody from outside the United States.

I want to be fair about this. A roundtable is not a rulemaking and it is not obliged to be representative. The Board's stated purpose was to hear from payments and technology industry representatives, and that is exactly what it did. Nobody was excluded from a decision, because no decision was taken.

But "no decision was taken" is doing a lot of work in that sentence, and anyone who has ever sat in a standards meeting knows why.

Decisions in payments are rarely taken in the room where they appear to be taken. They are taken in the accumulation of rooms: the shared vocabulary that forms, the problems that come to be treated as the real problems, the solutions that get described often enough to become the obvious ones. By the time something reaches a rulemaking with a comment file, the shape of the answer has usually been fixed for a year by people who were in rooms like this one.

That is not a conspiracy. It is how technical policy actually gets made, in every jurisdiction, in every industry. It is the normal and probably unavoidable mechanics of regulating something complicated.

It is also precisely the mechanism that the Fed's stated doctrine leans on hardest.

The doctrine, nine months on

In August 2025, Christopher Waller told a conference in Wyoming that private-sector innovation should be the default model, that Federal Reserve-built platforms should be used only rarely, and that the Fed's role was to provide core settlement, convene the industry around standards, do technical research, and engage with innovators. He said the Fed could benefit from further engagement with industry innovators, and that it was working on ways to do so.

This is what that looks like nine months later. Waller is in the room. So is the Governor who has been the Board's most prominent voice on consumer protection and bank regulation, and so is a third. They are listening to Visa, Mastercard, Stripe, Amazon, Google and a startup that wants to build agent payments.

The doctrine is being executed exactly as described. I said in August that the most important sentence in that speech was about who should build, and this meeting is the proof: the Fed is not building an agentic payments framework. It is listening to the people who are.

What I want to flag is the governance consequence, which I do not think is well understood.

A rulemaking is a terrible, slow, adversarial process with one enormous virtue: it produces a public record. The OCC's stablecoin proposal this year asked 211 questions and opened a comment file. Anyone — a bank in Panamá, a consumer group, an academic, a competitor — can read the questions, file an answer, and see what everyone else filed. The reasoning is contestable because it is written down.

Convening produces a guest list.

I am not accusing the Board of hiding anything; publishing the roster at all is more transparency than most convening bodies offer, and the Fed deserves credit for doing it. But the asymmetry is structural. If your governance model is convening, then participation is the only form of influence, and participation is by invitation.

From a Panamá desk

Here is what that means for us, stated plainly.

Agentic commerce will arrive in this region as a finished product. The rules about what an agent may do, how it identifies itself, who is liable when it buys the wrong thing, and whether a merchant's edge lets it through at all — those questions are being worked out now, in English, in rooms in Washington, among firms that are mostly American.

We will not be consulted, and I do not say that bitterly. There is no mechanism by which a Panamanian processor would be in that room, and no reason the Federal Reserve Board should invite one. The Fed's mandate is domestic. Our dependence on its decisions is not.

So the only available strategy is the one small markets always end up with: read everything, early, and arrive at the finished product already knowing what it is.

Concretely, for anyone running payments in this region, three things follow.

Watch the rosters, not just the rulemakings. A guest list published in May tells you what the agenda will be in the following year, and it is free. It is the cheapest piece of market intelligence available and almost nobody in this region reads it.

Assume agentic commerce reaches you first as a liability question, not a technology question. Your institution will be asked to authorise transactions initiated by something that is not a person, under rules written elsewhere, before anyone has told your risk committee what an agent is. Have an answer before that conversation starts.

And look at your own edge. If a European or North American agent tries to buy from a merchant you acquire, does your fraud stack treat it as a bot? Today, probably yes, and correctly. Tomorrow that may be the difference between a merchant who sells internationally and one who does not, and nobody will have told you the default changed.

My judgment

The Board convened the right room for the question it asked and the wrong room for the question that matters.

If the question is "how will agentic commerce be built," then Visa, Mastercard, Stripe, Adyen, Fiserv, Amazon, Google, Shopify, Coinbase, Plaid and Cloudflare are precisely the twenty people you need. They are the ones who will build it. There is no better-informed room in the world on that question.

If the question is "who bears the loss when it goes wrong," that room has a structural blind spot, because every participant in it is a firm that will be allocating losses rather than absorbing them. The parties who absorb the losses in payments are, in order: consumers who do not notice, small merchants who cannot negotiate, and small banks who cannot refuse. None of the three had a seat.

I think that will produce a framework that is technically excellent and distributionally predictable. The agent will work. The liability will settle on whoever has the least power to push it elsewhere. We have watched this exact sequence with card-not-present fraud, with 3-D Secure, with every liability shift the industry has engineered in thirty years, and the pattern has never once broken.

My second judgment is about Cloudflare, and I hold it with more conviction than is probably warranted by a guest list. I think the gatekeeping of agents at the web edge is the most underpriced risk in agentic commerce. The payments industry is arguing about credentials and liability while the actual admission control sits with three or four infrastructure companies and a default setting. If agentic commerce consolidates, it may consolidate not because a network won but because an edge provider decided which agents are allowed through and on what terms. That is a market structure question and nobody is treating it as one.

The third is about the document itself, and it is the reason I wrote this. The Federal Reserve Board published a guest list and one sentence, and that is more than most convening produces. It is still a guest list and one sentence. If convening is going to be the governance model for the biggest change in commerce since the browser, then the people doing the convening should publish more: an agenda, the questions asked, the disagreements. Not minutes — nobody would speak freely — but something a person outside the room can argue with.

Because right now the only thing I can argue with is who was invited, and I have just spent two thousand words doing exactly that, which rather proves the point.

Three questions for the people who own this

1. Does your institution have a written position on what an AI agent is, for authorisation purposes? Not a strategy deck — a definition your risk committee has approved and your authorisation logic could act on. If not, who is drafting it and by when?

2. If an agent attempts a purchase at a merchant you acquire, what does your fraud stack do today? Has anyone actually tested it, or are we assuming? And who owns the decision to change that default?

3. Which bodies outside your country are currently shaping the rules you will have to implement, and who in your organisation reads their published material each month? If the answer is nobody, that is a role, not an accident — whose job is it going to be?

Reply to this email with a line or two. The sharpest answers will shape the next issue.

Sources

Board of Governors of the Federal Reserve System, 26 May 2026 — published record of a Payments Innovation Roundtable meeting between Board members and staff of the Federal Reserve System and payments and technology industry representatives. Source for the date; for the stated purpose, which was the opportunities and challenges of artificial intelligence in payments with a particular focus on agentic commerce; and for the full list of participants. Board participants recorded: Governor Christopher J. Waller, Governor Michael S. Barr, Governor Lisa D. Cook, and staff including Sonja Danburg, Micah Gertzog, Michael Horowitz, Hannah Kim, David Mills, Alex Sproveri and Kirstin Wells. Reserve Bank participants recorded: President Thomas Barkin and Ghada Ijam of the Federal Reserve Bank of Richmond, and Mark Gould of the Federal Reserve Bank of San Francisco. Outside participants recorded: Karan Katyal (Adyen), Geoffrey Weien (Amazon), Scott Mullins (Amazon Web Services), BC Krishna (Centime), Vincent Voci (Cloudflare), Alec Lovett (Coinbase), Pravesh Rijal (Cross River Bank), Takis Georgakopoulos (Fiserv), Erin McCune (Forte Fintech), Nikhil Sharma (Google), Jackie Reses (Lead Bank), Mike Sulla (Mastercard), Serge Elkiner (Paze), Brian Dammeir (Plaid), Mani Fazeli (Shopify), Amir Sarhangi (Skyfire), Jonah Crane (Stripe), Alexia Pastré (Taktile) and Jalpesh Chitalia (Visa).

Christopher J. Waller, Federal Reserve Board, 20 August 2025 — "Technological Advancements in Payments," Wyoming Blockchain Symposium. Source for the doctrine described in this essay: that private-sector innovation should be the default model, that Fed-built platforms should be used only rarely, that the Board's role includes convening the industry around standards, and that the Fed could benefit from further engagement with innovators in industry and was working on ways to do so.

On what the record does not contain. The Board published no agenda, no discussion points, no participant positions and no minutes — only the summary sentence and the roster. Everything in this essay about what the meeting signifies is inference from who was invited and from the Board's own stated doctrine. I do not know what anyone said. I have not spoken to any participant. Where I characterise a firm's relevance — Cloudflare at the web edge, Skyfire in agent payments, Cross River and Lead as sponsor banks, Taktile in decisioning, Paze as a bank-owned wallet — those are my descriptions of what those companies do, not the Board's.

A note on what is mine. That the roster is the document; that the absence of consumer, merchant-association, community-bank and non-US voices will show up later as a predictable allocation of losses; that web-edge bot policy is the underpriced chokepoint in agentic commerce; and that convening as a governance model produces less contestable reasoning than rulemaking — those are arguments, and I would be glad to be wrong about the second one. Nothing here reflects the position, portfolio, clients or vendors of my employer.