The Third Principal
Three years ago I wrote that your digital identity and your financial identity were, in our modern age, one and the same — and that whoever solved the convergence of the two would hold the key to how we interact with the digital world. I called it Wallet Centrism. The thesis has aged well. Coinbase built settlement rails on exactly this premise. Visa launched a programme to let software transact on your behalf. Apple Pay quietly became the "login with your bank" thought experiment I floated, executed by a different mechanism than anyone expected. The convergence happened.
What I didn't predict — what I couldn't have predicted, sat in 2023 — is that a third identity would arrive and quietly break the entire model everyone was building. Not the human's digital identity. Not their financial one. The agent's.
And the agent's arrival doesn't merely extend the wallet-centric thesis. It proves it was right, by exposing the thing that was always implicit in it and that none of us had the vocabulary to name.
What the Essay Got Right, and Why
Wallet Centrism argued that financial identity should be digital identity — that the wallet was the convergence point where these two long-separated things would finally collapse into one. I framed this as a user experience problem. Seed phrases were unfit for purpose. Login should feel like login. The contents of your wallet should tailor your experience across every app you touch, projected from a single identity rather than fragmented across a dozen app-generated key-pairs.
All of that was correct. It was also, I now realise, the shallow reading of my own argument.
The deeper claim was sitting underneath the whole time, and it took the agent to drag it into the light. Authentication answers who you are. The wallet was never really about authentication. A wallet answers a different and more powerful question: what are you permitted to do? Logging in proves identity. Paying proves authority. The wallet was always an authorisation surface wearing the costume of an authentication one — and when humans were the only actors in the system, the distinction simply didn't matter. You were present. You authenticated and you authorised in the same breath, at the same keyboard, in the same instant. The two collapsed into a single act, and so nobody bothered to separate them.
That collapse was a luxury of presence. And presence is exactly what the agentic era takes away.
The Assumption Everyone Built On
Here is a fact about every identity system designed between roughly 2015 and 2024: each one assumed that the entity holding the identity was the same entity taking the action.
OAuth assumed it. FIDO assumed it. Passkeys assumed it. The entire edifice of modern web authentication rests on a single load-bearing premise — the person at the keyboard is the person who authenticated — and that premise was so obviously true for so long that it was never written down as an assumption at all. It was just the water everyone swam in.
The agent shatters it completely.
When a piece of software goes out into the world to book your travel, rebalance your portfolio, or buy the thing you asked it to buy, the entity executing the action is categorically not the entity that authorised it. You authorised. The agent executes. The person who said "yes" is asleep, or in a meeting, or three time-zones away and entirely absent from the transaction taking place in their name with delegated authority. The continuous-presence assumption — the one holding up the whole stack — is simply gone.
This is not a small crack. It is a foundational one. And it cannot be patched by the existing infrastructure, because the existing infrastructure was built on the wrong primitive. It was built on presence. What the moment demands is a primitive built on delegation — and those are not the same thing dressed differently. They are different things entirely.
Sessions Were Built for People Who Are There
Consider how the web actually works today. You authenticate, you receive a session token (read: a temporary pass that says "this person is logged in right now"), and you act within the bounds of that session. The session assumes you are present and continuous — that you are here, doing things, for as long as the token lives.
Agents do not have sessions. Agents have mandates.
What an agent carries into the world is something genuinely new in the architecture of the web: a scoped, cryptographically signed, time-bound, revocable authorisation that travels with it independently of any session — what at Sumvin we call a purchase intent, or PINT. It is not a session token; a session says "this person is here." It is not an API key; an API key says "this software is allowed to call this endpoint." A purchase intent says something both narrower and far more powerful — this specific human has authorised this specific category of action, within these specific bounds, for this specific window of time, and you can verify that cryptographically without the human being anywhere near you.
This is the architectural primitive my 2023 self was reaching for and couldn't quite grasp. When I wrote that I wanted "one identity which I want to project into all of my digital experiences," I was imagining a human doing the projecting — me, present, carrying my identity from app to app. I had the projection right. What I missed is that the projection would one day need to happen without the projector being in the room. The wallet was always an instrument of delegated authority. We only see it clearly now because, for the first time, the one holding the instrument isn't the one who owns it.
When that becomes the norm — when authority can travel safely without its owner — something fundamental shifts in what it means to act online. You stop having to be present for your own digital life. The web stops being a place you visit and becomes a set of standing instructions the world executes faithfully on your behalf. That is a far larger change than "AI does your shopping." That is the difference between operating a machine and employing one.
A Category Error, Repeated at Scale
Survey the landscape of companies attacking agentic commerce right now and you will find a great deal of genuine talent, real engineering, and serious money. You will also find, almost universally, the same mistake.
There is a whole cohort building what's come to be called "know your agent" — handing agents their own verifiable identities, their own credentials, their own reputation scores and verification lanes. There is a parallel cohort building agent wallets — handing agents their own balances, their own decentralised identifiers, their own email addresses and phone numbers and virtual cards. A SIM card for agents. Puppet identities, provisioned by the month.
Every one of these started from the agent and tried to build identity backwards towards the human. And every one of them has committed the same category error, just at different layers of the stack.
The agent is not an economic actor. The agent is a projection — the mechanism by which a human's authority reaches into a context the human is absent from. Giving an agent its own identity, its own wallet, its own reputation, is like issuing a passport to a person's shadow and then wondering why the border guard is confused. The identity that matters has never been the agent's. It is the human's. The agent's so-called "identity" is nothing more than the scope of what it has been permitted to do on that human's behalf — a permission slip, not a person.
The closest anyone comes to getting the sequencing right is World, with its proof-of-human approach — beginning, correctly, from the premise that there must be a real person behind every transaction. That is the right starting point. The problem is that it stops at uniqueness. Proving "this is a genuine, singular human" is necessary and nowhere near sufficient. The question the agentic era actually asks is not "is there a real person?" but "did this real person authorise this action within these bounds, and can I prove it, right now, without them here?" Proof of personhood is the foundation. Delegated, scoped, revocable authority is the building. Almost everyone has poured the foundation and called it a house.
There is, to be fair, a small handful reasoning from the right end — beginning with human authority and building the agent's permissions downward from it, carrying signed mandates across unified rails. Most of them are coming at it from enterprise procurement rather than consumer finance, which is a different bet with a different shape. That they exist at all is the strongest evidence I can offer that the sequencing argument is correct: the serious players, the ones who'll still be standing in five years, are quietly converging on it from opposite directions.
The Loop, Now That It Runs Without You
In Wallet Centrism I described a loop. Users arrive at an app, an identity is created for them, they carry that identity to the next app, and apps begin to build around the identity because that is where the users — and crucially, their money — already are. The cold-start problem was getting users to create wallets. The network effect was apps choosing to build from a wallet because the users were already inside it.
The agentic loop is the same shape and an order of magnitude more powerful, because the user is no longer the one walking through the door.
The agent enters services carrying the human's scoped authority. It transacts within its bounds. And the services — the merchants — begin to build around the credential, because that is now where the authorised purchasing power lives. The old game was attracting users. Merchants accepted certain payment methods to capture the people inside those systems. The new game is being discoverable by agents. A merchant no longer needs to win the user's attention; the user isn't there to give it. The merchant needs to be visible, parseable, and acceptable to the software the user has dispatched in their stead.
So the cold-start problem transforms. It is no longer "get humans to create wallets." It is "get merchants to accept purchase intents." And the moment enough of them do, the loop closes with a force the human-driven version never had: every merchant that verifies a credential makes every issued credential more valuable, and every issued credential makes verification more attractive to the next merchant. Each verifier raises the worth of every issuer. Each issuer deepens the pull on every verifier. The flywheel doesn't spin faster — it spins on a substrate that doesn't get tired, doesn't churn, doesn't need to be re-acquired every quarter.
Once that takes hold, the texture of commerce itself changes. The frantic competition for human attention — the ads, the funnels, the dark patterns engineered to keep you clicking — loses its grip, because attention stops being the scarce resource that the whole edifice was built to capture. What becomes scarce instead is trust: the merchant's standing to be accepted by a credential, the issuer's standing to vouch for what its users have authorised. We spent two decades building an internet that fought for eyeballs. We are about to build one that competes for the right to be trusted with delegated authority. That is a healthier internet, and a stranger one, and it is coming whether the attention-merchants are ready for it or not.
Identity Is the Whole Game
My 2023 essay closed on convergence — digital and financial identity merging into one. The honest follow-up has to close somewhere harder, because the agent forces a harder conclusion.
In the agentic era, the identity layer is not a feature of the product. It is the product. Everything else — the wallets, the cards, the on and off ramps, the budgets, the commerce — is downstream of getting identity right. It is a consequence, not a component. Whoever owns the credential that agents carry out into the world owns the relationship between human beings and their increasingly autonomous financial lives. And that is not a fintech company in any sense we'd recognise. That is infrastructure for the next era of economic activity — the substrate on which an entire generation of software will learn to act on our behalf.
My conclusion to the Wallet Centrism insisted that people deserved to project themselves digitally in the way they were supposed to, unbounded by the wrong consumer choice of card issuer. That was true, and it was small. Here is the larger version. We are about to hand pieces of software the authority to act for us, at scale, in the most consequential domain we have — our money. The only thing standing between that future and a catastrophe of misplaced trust is ensuring the authority those agents carry is scoped, signed, revocable, and provably ours. It’s imperative that we get this primitive right to ensure the whole agentic economy can be verifiably trusted to act on our behalf.
The convergence I wrote about three years ago was the beginning of the story. The third principal — the agent, acting in our name while we are elsewhere — is where it actually gets interesting. The infrastructure that makes its authority safe is the only thing worth building. So let's go and build it properly, before someone builds it badly and we all have to live inside the consequences.