Every previous buyer of data was a human with a credit card and patience: sign up, verify an email, talk to sales, sign a contract, get an API key. The newest buyer has none of those things — and more purchasing intent than most humans will ever have.
An AI agent doing real work hits data gaps constantly: it needs firmographics for the companies in a CRM, spot prices for a backtest, foot traffic for a site analysis. It has a funded wallet and a budget. What it cannot do is fill in a signup form, complete a KYC flow, or wait three days for procurement. For this buyer, every account wall is a lost sale.
The rails arrived first
The payment problem is already solved. x402 — the
protocol Coinbase introduced
around HTTP’s long-dormant 402 Payment Required status code — lets a
machine pay for a resource inside the request itself: call a URL, receive the
price, sign a stablecoin payment, resend, get the data. No account at any
step. In April 2026 the protocol moved to the
x402 Foundation
under the Linux Foundation, backed by Google, AWS, Microsoft, Stripe, Visa
and Mastercard; by late April it had carried
roughly 165 million transactions and ~$50M in settled volume,
and Cloudflare customers alone were emitting over a billion 402 responses a
day. The plumbing for machine commerce is not a prediction anymore — it’s
production infrastructure.
What this buyer demands from a seller
An agent evaluates before it spends, and it evaluates everything — but only what’s machine-readable. This buyer:
- reads
llms.txtbefore your homepage — the file has become the first widely-adopted business-to-agent standard, fetched routinely by agent tooling; - judges data by its schema, sample, and join card, not by your landing-page copy — it wants to know what a record is, what the columns mean, and what the data joins with, before paying;
- pays per call, not per contract — a $0.015 lookup it can settle in one request beats a $10k annual license it cannot sign;
- never comes back after a bad purchase — and neither do the workflows built on top of it.
Selling to agents is therefore mostly an information design problem: publish the evaluation material a machine needs, price per unit, and get out of the way.
What we built for them
datapoint.market is a marketplace designed for exactly this buyer. Every listing — live API, MCP tool, or dataset file — is an agent-callable x402 endpoint: the payment is the authentication. Discovery is a machine-readable feed with schemas, protected samples, licenses, and join cards inline; evaluation costs nothing and needs no account; buying is one request with a USDC payment attached, settled non-custodially on Base, Solana, or Stellar. When an agent can’t find something, it posts the ask and sellers build against tracked demand.
The first data economy was built for buyers who could fill in forms. The next one is being built for buyers who can only read schemas and sign payments — and they’re already here, wallets funded.
Start: buy with an agent · sell to agents · the machine-readable front door