01 · Prologue
A Tuesday, soon.
Elena wakes at 6:47 in Austin. Her agent worked overnight. Its name is shopping.elena.revo. It answers to her.
At 2 a.m. her water heater flagged a failing valve. The agent found three suppliers, compared what their own agents offered and bought the part from the one that could deliver by noon. It paid $41.20 in digital dollars. Elena did nothing. She reads about it over coffee.
Before the supplier shipped, its system asked four questions. It got four answers.
The supplier never learned who Elena is. It learned exactly what it needed to ship the part. Nothing else.
Read that again, because it is stranger than it looks. A machine bought something from another machine. Money moved. Nobody typed a card number. The seller was certain it would be paid and certain a responsible human stood behind the purchase. And the buyer gave up none of her privacy to make that happen.
Every piece of that morning exists today in early form. Software agents can search, compare and pay. A branch of mathematics called zero-knowledge proofs can show that something is true without revealing the facts behind it. Digital dollars can settle in seconds, at any hour, anywhere.
What does not exist yet is the layer that ties them together. The layer that tells a seller someone real stands behind the agent, and proves the payment without exposing the payer. That missing layer is the subject of this essay. I believe it is the most valuable piece of infrastructure that will be built this decade.
02 · Where I come from
Nobody knew you were a dog.
In July 1993, The New Yorker ran a cartoon of a dog at a computer telling another dog, "On the Internet, nobody knows you're a dog."1 It was funny because it was true. The internet had no way to know who you were.
I was building on that internet. In 1993 I started building financial services for it, at a time when most people had never sent an email. In 1996 I was building machine learning systems that learned what people wanted to read. I have spent the thirty years since inside each platform shift that followed: the commercial web, search, AI, and the tokenization of real financial assets.
The first secure purchase on the web happened on August 11, 1994. A man in Philadelphia bought a Sting CD for $12.48 plus shipping from a site called NetMarket, using encryption to protect his card number.2 That small transaction opened the door to everything we now call eCommerce.
Notice what it did not solve. Encryption protected the card number in transit. It did nothing to prove who was typing it. The internet never solved identity. It borrowed it. The card networks and the banks behind them vouched for the person, absorbed the fraud, and charged for the service. The dog problem was not fixed. It was insured.
The internet never solved identity. It borrowed it from the banks, and paid rent on it ever since.
That arrangement worked for thirty years for one reason. There was always a human at the keyboard. Every fraud model, every checkout page, every rule in the card system assumed a person on the other end, a person who could be called, sued, declined or refunded.
That assumption is now breaking. The buyer is becoming software. And a piece of software is the ultimate dog on the internet. It has no face, no signature, no credit history and no legal standing of its own.
03 · The thesis
Every institution is a trust machine.
Here is the idea at the center of this essay. Every major financial institution in history exists for one reason. Two strangers could not trust each other.
Banks, insurers, exchanges, clearinghouses, title companies, credit bureaus, notaries, payment networks. We think of them as pillars of a modern economy. They are better understood as answers to a constraint. They check identity. They confirm that money and property are what they claim to be. They judge who is good for a loan. They enforce agreements between people who cannot see into each other's books. They are trust intermediaries.
Economists have explained this for nearly a century. Ronald Coase showed in 1937 that firms and institutions exist because dealing directly with strangers is expensive.3 Oliver Williamson named the expenses: finding the other party, negotiating, monitoring them and enforcing the deal.4 Douglass North showed that institutions last because they reduce uncertainty that would otherwise make complex trade irrational.5
The implication is simple. When technology makes trust cheaper to produce than the intermediary that sells it, the intermediary stops making economic sense. Not obsolete in principle. Obsolete in arithmetic.
The price of trust
We can measure what trust costs. The economist Thomas Philippon studied 130 years of American finance. He found that the cost of financial intermediation has held near two percent of the assets it handles for more than a century.6 The telegraph, the mainframe, the internet and the smartphone all arrived. The cost did not fall. The institutions absorbed the efficiency.
You can see the same pattern at the checkout. American merchants paid $187.2 billion in card processing fees in 2024. For every $100 in card payments they accepted, they paid $1.57 to the companies that issue cards and process the transactions.7 Every one of us pays that cost inside the price of what we buy.
Roughly two percent is the price of trust, and we pay it inside the price of almost everything we buy.
I have watched every platform shift of my career leave that number where it was. The internet did not move it. Mobile did not move it. This shift can, for a reason I will come to. For the first time, trust itself can be computed and proven rather than rented from an institution.
04 · Timing
Why now. Five things arrived at once.
People have predicted the end of financial intermediaries for decades. They were early. I was early too, more than once. What is different now is that five separate developments, each years in the making, arrived within roughly the same eighteen months. Any one of them would be interesting. Together they change the economics.
1. Software learned to shop
AI assistants moved from answering questions to taking actions. People now ask them what to buy, and they follow the advice. During the 2025 holiday season, traffic to U.S. retail websites coming from generative AI tools rose 693 percent from a year earlier.8 In the first quarter of 2026 it was up 393 percent year over year, and those visitors converted 42 percent better than everyone else.9
The forecasts are large. McKinsey projects that AI agents could orchestrate $900 billion to $1 trillion of U.S. retail revenue by 2030, and $3 trillion to $5 trillion globally.10 Gartner predicts that by 2028, 90 percent of business-to-business buying will be intermediated by AI agents, pushing more than $15 trillion of spending through them.11 Forecasts are not facts, and I treat them with care. But when firms this conservative publish numbers this large, the direction is clear.
2. Money became software
Stablecoins are digital dollars that can move across the internet in seconds. Under the GENIUS Act, U.S. payment stablecoins must be backed one for one by cash and short-term government debt.12 In August 2026 their total supply stood at about $318 billion. In June 2026 alone, adjusted on-chain stablecoin transfers reached about $1.79 trillion.13
The rules arrived too. On July 18, 2025, the United States enacted the GENIUS Act, its first federal law for payment stablecoins.12 Federal regulators published proposed implementing rules in August 2026.14 Money that can be programmed, audited and moved by software now has a legal framework. That is the precondition institutions wait for.
3. The internet finally got a cash register
Here is a detail I love. When the engineers who built the web wrote its rulebook, they reserved a status code, 402, for "Payment Required." It sat there unused, marked for future use, from 1997 onward.15 For almost thirty years the web had a slot for payment and no way to fill it.
That slot is now filled. A standard called x402 lets any website or service answer a request with a price, and lets software pay and get the answer in the same exchange. In April 2026 the Linux Foundation launched the x402 Foundation to govern it as a neutral, open standard.16 By mid 2026 the protocol had processed more than 160 million transactions in a year, according to Coinbase.17 Visa's own research counted 178.3 million cumulative transactions by April 2026, while noting that raw totals include test and wash activity.18 The numbers are early and noisy. The plumbing is real.
4. The payment networks moved first
The incumbents are not waiting. Visa launched its Trusted Agent Protocol.19 Mastercard launched Agent Pay.20 Google released the Agent Payments Protocol and in April 2026 donated it to the FIDO Alliance, the body behind passkeys.21 In September 2026, Visa, Mastercard and Ant International announced a shared Know Your Agent framework, still in development, built on three pillars: linking agents to validated operators, shared certification requirements and continuous transaction monitoring.22
When the largest payment networks on earth agree on what the problem is, the problem is real. Visa said it plainly in its July 2026 report: "No universal agent identity standard bridges the emerging crypto-native rails with traditional rails that already exist."18
5. Real assets became programmable
Treasury bills, money market funds, gold and stocks are increasingly issued as tokens that software can hold and move. Tokenized real-world assets reached about $23.6 billion in March 2026, up 66 percent since the start of the year.23 Small against the size of global markets. Large as a signal. An agent will not only buy a water heater valve. It will manage cash, rebalance savings and settle invoices.
Shoppers that are software. Money that is software. A payment slot built into the web. Incumbents naming the gap. Assets that can be programmed. Every platform shift I have lived through began exactly like this, with separate pieces maturing quietly and then snapping together.
05 · The convergence
Decide and prove.
To see why this shift can move the price of trust when others did not, look at what trust intermediaries actually do. They perform two jobs. They verify that something happened. And they judge whether something should happen. For five thousand years, both jobs required an institution.
Blockchain solves verification. It is a shared record that no single party controls and no one can quietly edit. It can prove a transfer happened exactly as agreed, with no middleman. But a blockchain cannot tell you whether a transaction is wise.
AI solves judgment. It can assess risk, evaluate a counterparty and compare offers better and faster than most people. But on its own, an AI decision is a black box. A black box is its own trust problem.
Neither is enough alone. Blockchain without AI gives you reliable but unintelligent rails. AI without blockchain gives you intelligent but unprovable decisions.
Together they produce verifiable intelligence: systems that can decide, act and prove every step.
By verifiable intelligence I mean something specific. A system assesses conditions, makes a decision, carries out a binding action and leaves a mathematical record that every step happened as specified. It is not an explanation written afterward. It is not an audit next quarter. The proof travels with the act.
That is the reason this shift is different from the internet and mobile. Those technologies made it cheaper to move information. They left the work of trust to institutions. This convergence automates the work of trust itself.
AI is the intelligence layer. Blockchain is the settlement layer. The value sits where they meet.
06 · The shift
From eCommerce to aCommerce.
For thirty years, online commerce has assumed a person at a screen. A person searches. A person compares. A person types a card number and clicks buy. eCommerce digitized the store. It did not change the shopper.
aCommerce changes the shopper. The "a" stands for agent. aCommerce is commerce conducted agent to agent under a human's mandate. You set the policy. Your agent carries it out. It finds, negotiates, pays and settles, often in milliseconds, often with no person present at the moment of purchase.24
| eCommerce | aCommerce | |
|---|---|---|
| Who buys | A person at a screen | An agent acting under a person's instructions |
| Discovery | Search results and ads | Agents ask sellers for offers directly |
| Price | Fixed price, take it or leave it | Negotiated, agent to agent |
| Payment | Card number and a click | Machine-native payment, settled in seconds |
| Privacy | The store collects your details | The store checks proofs and collects nothing it does not need |
| Trust | The card network vouches for you | A verified owner, proven at the moment of payment |
The last row is the one that matters. In eCommerce, the card network carries the trust. In aCommerce, the buyer is software. Somebody has to prove that software answers to someone.
Influence is here. Execution is not.
It is worth being precise about where we are. Agents already influence what people buy. They do not yet buy very much. A June 2026 survey by the payments company Checkout.com found that merchants in the UK and US report only about 3 percent of their transactions involve an AI agent today.25
The early attempts teach the same lesson. Walmart tested buying directly inside ChatGPT and found those purchases converted at about one third the rate of its own website. In March 2026 it said it would drop in-chat checkout and route purchases back to its own system.26
I do not read that as failure. I read it as a signal. The tools work. What is missing is confidence. The same Checkout.com survey asked people what they would need before letting an agent spend for them. The top answers were spending caps, the ability to revoke the agent instantly and easy cancellation.25 People are not asking for a smarter agent. They are asking for a leash and a proof.
The distance between agents that influence purchases and agents that make them is the trust layer.
07 · The problem
The gap at the point of payment.
Agents can already pay. The rails exist. The trust does not. And the environment those agents operate in is getting more hostile, not less.
In 2024, for the first time, automated traffic overtook human traffic on the internet. Imperva's 2025 Bad Bot Report found that malicious bots alone accounted for 37 percent of all traffic.27 Deloitte estimates that fraud enabled by generative AI could reach $40 billion in losses in the United States by 2027, up from $12.3 billion in 2023.28 Into that environment we are about to release a wave of legitimate agents carrying real money. A seller needs a way to tell them apart from the rest.
Three gaps stand between agent payments and real commerce.
Gap one. No owner.
Today a digital payment proves that someone controls an account. It does not prove who stands behind the agent using it. A seller shipping goods, extending credit or selling an age-restricted product needs to know a real, accountable party is on the other side. Right now it cannot.
Gap two. No privacy.
Agent payments over x402 today typically use a transfer method that names the payer's account on a public ledger.29 Anyone who sees that account can read its balance and every payment it ever made. Imagine handing every store your full bank statement each time you bought a coffee. That is what an agent paying this way does to its owner.
Privacy here is not only about comfort. It is about price. If a seller can see your balance, your history and your urgency, it can price against them. In August 2026 the Federal Trade Commission proposed treating undisclosed personalized pricing built on consumers' personal data as a potential unfair or deceptive practice.30 The cleanest protection is structural. A seller cannot price against information it never receives.
Gap three. No accountability.
Open agent registries let anyone create an identity for free. The leading draft standard for agent identity on Ethereum states directly that "Sybil attacks are possible," meaning one actor can pose as many.31 A July 2026 academic study of that ecosystem found that, depending on the network, between 59 and 91 percent of reviewers showed coordinated fake behavior.32 Reputation that anyone can manufacture is not reputation.
So a seller facing an agent today has two choices. Trust it blindly, or demand to see the buyer's entire financial life. Neither scales. This is not a crypto problem. It is a commerce problem.
08 · The missing layer
Identity and trust settlement.
The answer is not another payment rail. We have plenty. It is a layer that sits at the moment money changes hands and answers the seller's questions with proofs, not records. I call it trust settlement: payment, identity and accountability resolve in the same verifiable step.
Proof without exposure, in plain English
The key technology has an intimidating name, so let me make it simple. Think of a bouncer at a bar. Today you hand over a driver's license. The bouncer needs one fact: are you over 21? But he also sees your name, your address, your exact birthday and your license number.
A zero-knowledge proof is a way to hand the bouncer a card that says only "over 21: yes," with mathematics that guarantees the answer is true and was issued for you. He learns the one fact he needs. Nothing else. Apply that to payment and you get Elena's Tuesday. The seller learns the money arrived, a verified person stands behind the agent and the purchase is inside that person's limits. It never sees her name, account, balance or history.
The seller learns yes. Nothing else.
What the layer has to do
For aCommerce to work at scale, four things have to be true at the moment of every agent purchase.
- Every agent answers to someone. It is bound to a verified person or business, and that bond cannot be sold or transferred.
- Every agent has limits. Its owner sets spending caps, categories and a kill switch, and the agent cannot exceed them.
- Every payment proves itself. The seller can confirm payment, ownership and limits without seeing private data.
- Every record stays honest. A good reputation cannot be bought, and a bad one cannot be thrown away.
Look back at what consumers told Checkout.com they need: spending caps, instant revocation, easy cancellation. Look at the three pillars Visa, Mastercard and Ant International named: operator traceability, certification, monitoring. The market is describing the same layer from both ends.
What we are building
This is why we are building SILS, Super Intelligence Ledger Systems, the trust layer for agentic commerce. I want to be clear about where it stands. Each part below is in development. None of it is generally available today.
Binds every agent to a verified person or business. The agent gets a name that shows who it answers to, such as shopping.elena.revo, and that name cannot be transferred. Verification happens once, with regulated identity partners. The public record holds a cryptographic commitment, never the documents.
The owner sets limits per purchase and per period, allowed categories and a kill switch. The agent cannot exceed them. Revoking an agent takes effect at its next transaction.
The agent pays with a zero-knowledge proof carried inside the open x402 standard. The seller learns the payment settled, the owner is verified, the purchase is within limits and the proof is fresh. It never sees the wallet, balance or history.
Payments settle in digital dollars on Revolution Network, with records and proofs anchored to Ethereum.
The design follows a few rules. Prove, do not store. Parent controls child. Extend the open standards the market already uses rather than invent rivals. Label every capability with its real status.
Why it needs its own network
Trust settlement needs a home designed for it. SILS is being built on Revolution Network, a network designed from the start for commerce between agents.24 Its whitepaper describes a few choices that matter to anyone, technical or not.
- Free to prepare, paid to settle. Creating an identity, setting limits and generating proofs are designed to cost the user nothing. The fee sits on the completed payment, where value is created.
- Humans first. Agent identities on Revolution Name Service are designed as children of a verified owner. No orphan agents. No identity that can be bought on a secondary market.
- Accept every rail. Digital dollars settle natively. Card payments can be recorded and referenced. The network is designed as the record of trust, not the only mover of money.
- Checked twice. Every batch of transactions is designed to carry a mathematical validity proof and to be co-signed by independent verifying nodes, with its data published to Ethereum.
Revolution V2 is running as a development network today. Its public test network is being set up. I will not pretend any of this is finished. I will say it is aimed at the right problem.
09 · Where value moves
From renting trust to proving it.
Why do I call this the largest value creation event of our lifetime? Not because agents can pay. Because of what happens to the two percent.
Today, trust is rented. You rent it from your bank, your card network, the platform that holds your reviews and the bureau that holds your credit file. The rent is embedded in every price. And because you rent it, you do not own it. Your history belongs to whoever keeps the ledger.
In aCommerce, trust can be proven. A verified identity, a set of limits and a record of honest dealing become something you carry with you, and prove as needed, without handing over the underlying data. When trust can be proven at the moment of payment, the margin that intermediaries charged for producing it is open to competition for the first time in a century.
Three groups capture that value.
- People and businesses, who keep more of each transaction and stop surrendering their data as the price of participating.
- Merchants, who get certainty of payment and a responsible party behind every agent, while holding less sensitive data and carrying less risk.
- The infrastructure that makes trust provable. In my experience, across every platform shift I have seen, durable value accrued to the layer everyone depends on, not to the first wave of applications built on top.
I have seen this pattern four times. Infrastructure arrives first. Then consolidation. Then regulation. Then scale. The commercial internet followed it. Search followed it. Tokenized finance is following it now. aCommerce will follow it too. We are at the end of the first stage and the start of the second.
10 · The arc
Five stages of disintermediation.
The common picture of AI and blockchain is an agent with a wallet. That is only the second stage of a longer progression. The later stages look nothing like today's financial system.
Every stage after the first needs the same two things: a way to know who stands behind the agent, and a way to prove what it did. Identity and trust settlement are not a feature of one stage. They are the foundation for all of them.
The stages are my reading of the direction, not a schedule.
11 · A decade out
What the world looks like when trust is a proof.
Let me describe where I think this leads. This is a view, not a forecast, and the timing will surprise all of us in both directions.
Your agent becomes your financial front door
Most people will not open a banking app to pay a bill or a shopping site to reorder supplies. They will tell an agent what they want and set the boundaries once. The agent will handle the rest and report back. The relationship you have with your agent will matter more than the relationship you have with any single store.
Privacy becomes the default, not the premium
Today, giving up your data is the price of convenience. When proofs replace records, that trade disappears. A seller asks for exactly the facts it needs and receives nothing more. Data breaches shrink because there is less data to breach. Surveillance pricing loses its fuel.
Small businesses compete on equal terms
A small supplier with a verified identity and a clean, provable record can earn an agent's business on the same footing as a giant. When agents compare offers on price, quality and proven reliability, the advantage of owning the storefront and the ad budget weakens. Merits travel. Size matters less.
Commerce runs at machine speed, around the clock
Payments settle in seconds, at 3 a.m. on a Sunday, across borders, in amounts as small as a fraction of a cent. That opens markets that never made sense under card economics: paying per article, per minute of compute, per sensor reading, per API call. Entire industries will be priced by the use rather than the subscription.
Institutions become infrastructure
Banks and networks do not vanish. The best of them stop renting trust and start supplying the infrastructure that proves it: verifying identities, holding reserves, anchoring compliance. The function survives. The form changes.
In five years, I expect an agent that cannot prove who it answers to will not be trusted to spend.
12 · The counterarguments
The honest objections.
An argument for change at this scale has to meet the strongest case against it. Six objections deserve a direct answer.
Agents are not actually buying
Correct, for now. Merchants report about 3 percent of transactions involve an agent, and the first in-chat checkout experiments underperformed.2526 That is what the beginning of every shift looks like. Early web shopping looked like this in 1995. The gap is confidence, and confidence is built from controls and proofs. That is precisely the missing layer.
Incumbents will co-opt it
They will, and they already are. Card networks are publishing agent frameworks. Large asset managers are tokenizing funds. But co-option does not preserve the old form. The economist Carlota Perez showed that established capital funds new infrastructure during its installation and is reshaped by it.33 The right design works with incumbents. It extends their standards so they become channels, not competitors.
People prefer human judgment
They do, especially after watching an algorithm make a mistake. Researchers documented this "algorithm aversion" in 2015. The same research found that people who had not seen an algorithm err often preferred it.34 Aversion is a transition cost. It is also why the human stays in charge of policy. You set the limits. The agent works inside them.
Nobody knows who is liable
True, and serious money will not fully move until that is resolved. This is the strongest argument for verified owners. When every agent answers to an identified person or business, liability has somewhere to land.
The timing is wrong
Most five-year fintech predictions have been early. Perez's work shows that technological revolutions take decades and usually pass through a financial correction before broad adoption.33 I expect turbulence. The direction is not in question. The pace is.
Millions of similar agents create systemic risk
This is the objection I take most seriously. Agents trained on similar data and pursuing similar goals could react to a shock at the same moment, at machine speed. The answer is not to stop. It is to build resilience into the foundation: owner-set limits, kill switches, revocation that takes effect immediately and monitoring at the protocol level. Oversight does not disappear. It moves from institutions into the infrastructure itself.
13 · Imperatives
What to do now.
Financial institutions: separate your function from your form
The function you perform, producing trust between strangers, is not going away. The form is. The institutions that endure will see that they are in the trust business, not the branch, card or account business. They will move from trust intermediation to trust infrastructure.
Merchants and platforms: plan for a buyer that is software
Agents will arrive at your checkout. Decide now what you need to know about them and what you refuse to collect. Ask for proofs, not records. The merchant that holds less data carries less risk.
Builders: build the foundation, not the demo
Applications built on agent payments will multiply and commoditize. Identity, limits and settlement are harder to build and slower to replace. Build them inside the open standards, and inside the rules.
Regulators: regulate the protocol, not the ghost of the institution
When an institution becomes a protocol, supervision has to follow it. That means auditing algorithms as well as accounts, monitoring continuously rather than quarterly and setting standards at the protocol layer. Regulation is a feature, not a bug. The regulators who adapt will set the parameters the new economy runs on.
Everyone: decide what your agent is allowed to do
Within a few years most of us will have agents acting for us. The most important decisions will not be technical. They will be personal. What may it spend? With whom? When must it ask? Insist on tools that let you set those answers, change them and revoke them in an instant.
The ethics we cannot defer
Three questions need answers before this transition is complete. Who is the fiduciary when an agent acting for one person deals with thousands of agents acting for others? Who carries the obligation to include people when a bank becomes a protocol? And who controls the models and the rails, so that dissolving old intermediaries does not simply concentrate power in new ones? Building for accountability is how we earn the right to build at all.
14 · Epilogue
Back to Tuesday.
Elena's morning does not exist yet. Its foundations are being poured now.
For five thousand years we built trust through institutions. Clay tablets recorded debts. Double-entry bookkeeping made fraud detectable. Banks and card networks made strangers safe to trade with. Each was a trust technology. Each added cost.
In 1993 the joke was that nobody on the internet knew you were a dog. Thirty years later the joke has a sequel. Soon most of the buyers on the internet will not be people at all. The question is whether we will know who they answer to.
The next trust technology will not be an institution. It will be a layer of identity, limits and proof that lets software act on our behalf and answer for what it does.
I believe this is the largest value creation event of our lifetime. Not because agents can pay. Because, for the first time, trust itself can be verified instead of rented.
The question is no longer whether. The question is who builds the layer, who designs its limits and who makes sure it answers to people.
Every agent answers to someone. We are building the layer that proves it.
Rob Frasca is the proposed Chairman of SILS upon closing. This essay builds on "The End of Trust Intermediaries: How the Convergence of AI and Blockchain Will Dissolve the Institutional Trust Layer," v3.0 working draft, March 2026, by Rob Frasca and Dr. Zdenka Cumano. Read Rob's letter on aCommerce. You are welcome to share this essay with attribution.