One word, three meanings.

When you hear “token”, what do you picture?

A coin? A password? The units an AI charges you for?

All three are right. That is the problem, and the opportunity.

Three meanings

Security. A token stands in for a permission. Your bank app’s login. A key card. A proof that you are over 18 that shows nothing else. The token says “this one may”, without giving away the secret behind it.

Money. A token stands in for a claim on something real. A share. A loan. A slice of a building. The token is a record you can transfer, faster and in smaller pieces than paper allows.

Intelligence. A token is a unit of what an AI reads and writes. Words are cut into pieces, and you pay by the piece.

Three meanings, built by different people, for different reasons. So why put them in one sentence?

They have started to meet

Picture an AI agent doing a job for your business. It calls a service. The service runs a model. The agent pays. To do that, one request must carry three things:

• Who may act: the agent’s permission, and the limit on it. That is security.
• What it may spend: a payment, up to a cap. That is money.
• What it used: the AI tokens it consumed. That is intelligence.

This is no longer a thought experiment. One open payment scheme for agents, x402’s “upto”, lets a seller set a maximum price, the agent sign for that cap, and the final charge match what was used. Its own docs give AI token generation as the example (x402 docs).

A spend cap, a compute meter and a signed permission, in one exchange. What does it mean when three kinds of trust travel together?

What stays the same

Here is the part most hype skips. A token does not change what you own. A deposit is still a deposit. A share is still a share. After its largest test, the Bank for International Settlements said tokenisation “does not alter the legal characterisation” of bank deposits or central bank reserves (BIS, 27 May 2026).

What changes is where the trust sits. It moves from a clerk to a ledger. From a signature to code. From “call me if it goes wrong” to rules that run on their own.

So the real question is never “is it on a blockchain?” It is: who checks this token, and what happens when they’re wrong?

Every meaning has its own checker

• A permission is only as good as whoever can revoke it.
• A claim is only as good as the law, the register and the money it settles in.
• An AI meter is only as good as the count behind it.

When the three meet, the checks don’t merge. They stack. A payment can prove the money was there and still not prove which human allowed the agent to spend it. That gap is where today’s fraud lives. In 2024 a finance worker at Arup’s Hong Kong office paid out about US$25 million after a video call in which every other face was a deepfake (Fortune).

The seams between the three are where things break. They are also where the value will be made.

What we think now

Tokens move the trust problem. They don’t remove it.

The winners will not be whoever issues the most tokens. They will be whoever engineers the seams best: which record wins, which money settles, which person said yes, and how fast that yes can be taken back.

For a business, that starts small. Know what you are trusting, who checks it, and who answers when it fails.

What we’re unsure of

• Who holds the off-switch? When an agent’s permission must be withdrawn across several companies at once, no standard yet says how.
• Who pays for an agent’s mistake? Today’s card rules were not written for an honest machine that buys the wrong thing.
• Which record wins? When a token and the old register disagree, the answer is still often “it depends”.
• How fast will Aotearoa move? New Zealand has no crypto-specific law yet, and the regulator has asked for one (NBR, March 2026).

What would change our mind

• A permission that can be proven and revoked across companies, in production, at scale.
• A clear legal answer, somewhere that matters, for who pays when an agent errs.
• Evidence that tokenized assets trade more easily than the assets underneath them, not just faster.

Your turn

Pick one thing you rely on every week: a login, a payment, an AI tool. Ask three questions:

• What is this token standing in for?
• Who checks it?
• If it’s wrong, who notices first?

If you can’t answer the third, you’ve found where to look next.

Read it with Tokenomics: pay the work, not the gate, on who a token pays.

Published 3 November 2026. A reflection on how systems are changing, not financial or legal advice.