Skip to content

The Journal · Agents

The market priced a coordination layer. It did not prove one.

September’s crypto and AI rally is a bet that blockchains might handle identity, permission and payment for software that spends. It is not evidence of AI revenue.

Grayscale’s artificial-intelligence crypto sector gained 54 per cent in September, against 24 per cent for the broader crypto market. NEAR rose 183 per cent, Venice’s VVV 70 per cent, World’s WLD 47 per cent, Bittensor’s TAO 37 per cent. Those are Grayscale’s figures. They describe a price.

The wrong sentence is that tokens went up, therefore blockchains power artificial intelligence. The right sentence is duller. Markets have started to price a possible coordination layer for an internet in which software buys things. Almost none of that commerce is visible as revenue.

Chapter 01 · 1 of 4

Models do not need a chain to think

Generating text, running inference and training a model are latency problems. Centralised infrastructure will keep winning most of them. A blockchain is a strange place to put a GPU cluster, and a rally does not change the physics.

The opening appears later, when a piece of software holds a wallet, hires compute, buys data, pays a supplier and has to explain itself to a stranger. The web is bad at that. It has accounts, not principals. It has sessions, not mandates.

Chapter 02 · 2 of 4

What the rally is actually pricing

AI produces intelligence. A ledger can hold identity, ownership, payment and the right to revoke them.

The missing pieces are ordinary once they are listed. A persistent identity a machine can read. A credential that can be checked and withdrawn. Programmable payment. A spend limit. An audit trail for an action nobody sat and clicked. A way for two firms that do not share a platform to settle. Incentives, if they are real, for data, compute, models and human review.

The networks in the September tape point at different parts of that list, and each has a question rather than a medal. Can NEAR become an operating layer for agentic commerce, or an L1 with a narrative. Can Bittensor’s incentives pay for useful models rather than for gaming the incentive. Can a proof-of-person system be trusted once the internet is full of synthetic actors. Can a privacy-first distribution business hold users and economics at once.

Chapter 03 · 3 of 4

The theme is verifiable action, not decentralised intelligence

Every firm that lets software act will need answers a chatbot cannot give. Which human or company authorised this. Is the agent allowed to spend. What data did it touch. Which policy governed the decision. Can the action be audited. Was the supplier real. Who gets paid when the agent buys inference or a real-world service.

That is a trust layer. It is not a token with an AI ticker. A business completes verification once and receives a credential it can reuse. It delegates a narrow authority. The agent can prove whom it represents and what it may do. Payment settles in stable-value money. A counterparty checks eligibility without receiving the whole data room. If the firm, the wallet or the agent changes status, the authority can be revoked the same day.

Chapter 04 · 4 of 4

A procurement agent with a leash

Picture a procurement agent for a UK construction company. It may source materials and negotiate inside a range. It may not release a payment unless the supplier holds a valid credential, the invoice matches the order, and the spend sits inside a mandate a person already approved. The leash is the product. The model is a component.

September priced the possibility. It did not pay anyone for having built it. The work is the mandate, the credential and the revocation, not a narrative coin issued to celebrate the chart.

All field notes

Illustrative commercial design only. This is not legal, tax, investment, employment or securities advice, and it is not an offer to the public, a solicitation to invest, or a promise of funding, equity or returns. Structures vary by venture and must be drafted for the incorporation jurisdiction by qualified startup, tax and securities counsel. Blockchain Lab does not claim to be a licensed investment manager, broker or bank on the basis of this website.