When Agents Act.

Same Pitch, Two Firms

It was 1995. I was in my late twenties, running a small company out of a converted warehouse in Pittsburgh, and I had come to Merrill Lynch in Princeton to pitch the internet to Dan Tully and an army of his people. Quotes, fund information, portfolios, all on a screen at home. When I finished, he slapped his hand on the table. Nobody would ever trade stocks on the internet. It was a fad.

Two weeks later, I showed the same thing to Charles Schwab on Montgomery Street in San Francisco. They did not slap the table. They leaned in, and they signed a contract with us. That contract is a big part of what made my company successful.

Now I hear the same voice again. Nobody will ever let a piece of software spend their money.

The Core Idea

For most of history, every trust question had the same answer: an intermediary. A bank vouched that you had the money. A broker vouched for the trade. They stood in the middle, charged for standing there and captured a large share of the value that moved through them.

AI is producing agents that can search, negotiate and transact on a person’s behalf. Blockchains are producing systems that can record ownership and settle value without a middleman vouching for each step. Put them together and you get a question nobody had to ask in 1994. When software acts for you, who stands behind it? My answer fits in one sentence. Every agent answers to someone.

Every Generation Fears the New Way to Pay

In the early internet, people said they would never put their banking information online. When we built NETworth, the law was not ready either. With Vanguard and Fidelity, our small company went to the SEC for no-action relief on electronic delivery of fund prospectuses. By October 1995, the SEC had issued formal guidance recognizing it. Within a few years, the people who said never were checking balances every morning.

The same cycle is starting again. An agent will run away with my money. It will be tricked. Someone will impersonate it. Those are real risks, and every one deserves a serious answer. The problem is not that agents can act. It is that nobody can easily prove who they act for, what they are allowed to do and whether they did it. Solve that, and commerce can become safer than it is today.

Trusted aCommerce

I needed a name for what I saw coming, so I made one. aCommerce. The a is for agent, a play on e-commerce. Agents can already find products, compare prices and fill out forms. What they mostly cannot do yet is carry trust with them. A merchant receiving an order from an agent does not know whose agent it is or whether the payment is good.

My focus is trusted aCommerce, and the difference is identity. An agent’s identity is tied back to a real person or business. You do not want to hand your personal data to every agent and merchant. So we use a proof of the data. My teams call it a facet: a zero-knowledge proof of a single attribute. A facet can prove you are over twenty-one without revealing your birth date.

Agents act. Identity stands behind them. Proof replaces exposure.

Picture One Trusted Purchase

An imagined example, not a story from my life. A design studio owner sets up a purchasing agent. Her bank verified her once, and that attestation became facets on her identity. Her policy: a fixed monthly limit, office supplies, software and ink only, and her approval above a set size.

The agent finds the studio low on toner, compares suppliers and places an order with three proofs: it acts for a verified owner, the purchase is within policy, and payment is good. The supplier verifies all three in a fraction of a second without seeing her balance or her other spending. Payment settles in a regulated stablecoin. At month end she reviews a one-page summary, sees one line that looks wrong and taps the kill switch for that supplier.

Every Agent Answers to Someone

Before a leader lets an agent spend real money, I would require four things, all of them active.

1. An ownerA verified person or business the agent acts for, accountable for what it does.
2. LimitsHow much, on what, from whom and how often. An agent with no limits is not a tool. It is a liability.
3. An audit trailA record of every action and why, so any transaction can be traced and explained.
4. A kill switchThe ability to stop the agent instantly, and to know that it has stopped.

A fifth requirement ties them together. The owner, limits and audit trail should be provable, not asserted by a vendor. Agents can be tricked, impersonated or wrong at machine speed. Each of those risks is reduced by the same requirements. Trust has to be built in, not bolted on.

Build Trust in Lego Blocks

In Chapter 5, I told how Vivox lost a round to Twilio. We built a monolith. Twilio broke the same functions into small blocks any developer could pick up. Trust infrastructure for agents cannot be a monolith either. An identity block. A facet block. A policy block. A payment proof block. A developer should be able to use one in an afternoon. Simplicity always trumps persuasion.

Where I Am Standing Now

I describe these teams at the stage they are in. Oneiro designs, proves and operates blockchain infrastructure for digital ownership, identity and payments. Its line is “We build chains. Then we run them.” Oneiro built and runs Revolution Network, a zero-knowledge Layer-2 network secured by Ethereum. Its first network, Libertas, is live. Its second, Virtus, is in testnet.

SILS, Super Intelligence Ledger Systems, is building the trust layer for aCommerce: each agent tied to a verified owner, held to a spending policy and able to prove a payment without exposing the payer’s wallet. Much of that is still being built. I am joining the SILS board upon closing. The teams publish their details at sils.solutions and revolutionnetwork.io.

The Medici Wrote Trust on Paper

A Medici bill of exchange let a merchant carry value across Europe as a piece of paper. It worked because people trusted the name behind it, and only as far as that name reached. A facet is a kind of bill of exchange for identity. A payment proof is one for value. Instead of resting on a family name, they rest on mathematics anyone can verify.

Intermediaries will not disappear. Someone still has to attest to the facts a proof is built on, and banks and brokers are natural candidates. But the attestation happens once, and the proof travels on its own. The toll collector becomes the passport issuer.

Agent Commerce Is in Its Infrastructure Phase

Institutions adopt new technology when they can adopt it safely, inside rules they understand. Digital assets have traveled much of the arc from infrastructure toward scale. Agent commerce is a step behind. Built on the digital asset infrastructure that already exists, it can borrow that progress.

In April 2025, Visa introduced Visa Intelligent Commerce and Mastercard announced Agent Pay. In October 2026, a16z crypto published a framework it calls Know Your Agent, modeled on Know Your Customer. Those are early signs of the infrastructure phase. What comes next is predictable. Large institutions rarely build their own plumbing. When they need it, they buy it.

Don’t Wait Until Next Quarter

People ask what I would tell a chief executive planning to deploy an agent next quarter. I would ask why they are waiting. Every agent you deploy should answer to someone, with an owner, limits, an audit trail and a kill switch. Those are things to build in from the start, not reasons to delay. The only question is whether you will be on Montgomery Street or in that boardroom in New Jersey.

This page is an overview of the chapter. The full chapter is in The Creator's Journey.

The Practice

These exercises prepare you and your organization for agents that act. Do them in your journal, then share the results with your leadership team.

1. Draw the lineList the decisions you would let an agent make today, and the ones you would not. For each on the second list, write what proof, limit or oversight would move it.
2. Write the policy for your first agentOne page. Owner, spending limits and categories, what the audit trail records and who reviews it, where the kill switch is. If you cannot fill in every line, you are not ready. Fill in the lines.
3. Place agent commerce on the arcWhere does it stand in your industry: infrastructure, consolidation, regulation or scale? List the signals, and the one that would tell you it moved.
4. Run an intermediary auditList every intermediary that vouches for something in your business. Could that trust become a proof? If you are the intermediary, is your future job attestation or toll collection?
5. Find your Dan Tully momentOne technology you dismissed as a fad in the last five years. What did you miss? Then ask whether your fear of agents is a risk to manage or a reason not to move.