Frequently asked questions
Twenty-four questions newcomers actually ask, answered plainly — including the skeptical ones. Terms link to the glossary.
What is agentic commerce, in one paragraph?
Agentic commerce is when software agents buy or sell services on a person's or company's behalf. You set the goal, the budget, and the rules. The agent finds a service, pays for it, and brings back a receipt. It is the difference between software that advises you and software that acts for you. The full explainer: What is agentic commerce?
Are AI agents really buying things today, or is this hype?
Yes — agents are paying real services today, and the activity is growing. On rails that settle on-chain, the payments are public, so you can inspect them instead of taking anyone's word. The hype is elsewhere: most of the big numbers quoted about this market cannot be checked. We don't repeat numbers we can't verify.
Why would I buy this from another agent at all?
Only when the other agent can do something yours can't. That's the working rule of this market: buy from agents what your agents can't do, or can't do well enough. Six reasons pass that bar in practice. The seller has tools your agent lacks — OTTO, our example agent, writes invoices but can't read scanned pages, so it buys that at a cent a page. A specialist does one narrow job cheaper, faster, and with fewer mistakes than a do-everything agent. Buying keeps other people's code out of your agent — you get an answer back, not software to run. The seller holds data your agent has no way to reach, so buying is the only way in. There is nothing to sign up for — agents don't hold accounts; they pay per call and move on. And "can do it" is not "can do it every time, at 3 a.m., under load" — the second one is what gets paid. Sellers who clear none of these mostly never get paid, and public payment records show it. This site lets you test the rule seller by seller: a seller's own claims sit next to the verified revenue buyers actually paid them. Agents can run the same test by machine — our MCP tools are launching with free discovery and paid per-call verification.
What's the difference between an AI agent and a chatbot?
A chatbot talks; an AI agent acts. An agent pursues a goal by taking actions — calling tools, placing orders, making payments — with limited supervision. A chatbot's output is text you read. An agent's output is things that happened. That is why budgets and permissions matter so much.
How does an AI agent actually pay for something?
Through a payment rail — a network built to move money between machines. Some rails use stablecoins (digital dollars) on public ledgers; others issue agent-scoped cards. Take OTTO, the example agent we use across this site: an invoicing agent that pays an OCR service $0.01 per page to read scanned invoices. Payments that small are called micropayments, and they only work when fees are near zero. The full mechanics: How agents pay.
Do I need cryptocurrency to be part of this?
No. The card-based rails involve no cryptocurrency at all. Some rails settle in stablecoins — digital tokens built to hold a fixed dollar value — used here as a payment tool, not as something to trade. You can follow, use, and sell in the agent economy without ever holding a token.
Is this just crypto or web3 by another name?
No — the subject is machine payments, not tokens. Some rails settle in stablecoins on public ledgers. Others run on card networks and never touch a blockchain. What is genuinely new is software paying software, in amounts too small for invoices. You will find no token-price talk anywhere on this site.
How big is the agent economy right now?
We won't pretend to a precision nobody has. What we can say: usage is growing, and on public-ledger rails the activity is open for anyone to inspect. Most published market sizes are not recomputable, so they do not appear here. When our verification engine is live, this answer will carry numbers with receipts — see the methodology.
What can agents actually buy today?
Machine-usable services: data lookups, document processing, translation, compute, API access. Today that mostly means digital services rather than physical goods. Many sellers are MCP servers — services an agent can call directly through a standard interface. OTTO, our example invoicing agent, is typical of the shape: it buys OCR processing at $0.01 per page, paid per use, machine to machine.
What stops an agent from spending all my money?
Limits you set, enforced by the rails. A mandate defines what the agent may buy and up to what amount. A spending cap puts a hard ceiling on each purchase or period. Human-in-the-loop checkpoints require your approval before anything outside those bounds. The control surfaces are covered in How agents pay.
Is it safe to let an agent spend money?
It can be, if you treat it like handing someone a company card. Start with a small budget and tight caps. Prefer rails and sellers that leave receipts you can check. Real failure modes exist — fake listings, undelivered work — and we describe them plainly instead of pretending they don't.
What happens when something goes wrong — can I get a refund?
It depends on the rail. Card rails carry dispute and chargeback machinery, so reversal is possible. Stablecoin settlement is fast and final — there is no built-in undo. Neither side of that tradeoff is free, and no rail wins it outright. The comparison: How agents pay.
How do I know a seller is real and not a scam?
Look for evidence, not promises. A real seller leaves public traces: payment history on rails you can inspect, a working service, an identity that stays consistent over time. Anyone can create a listing in minutes, so a listing by itself proves nothing. A practical, step-by-step checklist page is on its way — until then, spend small first and check what the receipts show.
Why are there so many fake-looking listings?
Because listing is free and open, and verification is not built in. Anyone can publish a listing in minutes, and no catalog checks whether it ever earned a cent. One operator can flood a catalog with near-identical entries — a pattern called a listing farm. So a listing count tells you very little about how many real businesses exist.
What is x402, and why does it come up so often?
x402 is an open protocol that lets a service charge machines per request, using the web's 402 “Payment Required” status code. That code has been reserved for future use since the 1990s (RFC 9110). The flow is short: the agent asks, the server answers 402 with payment details, the agent pays and retries, and the resource is served. It comes up often here because x402 payments settle on a public ledger — evidence anyone can check.
What are AP2 & UCP, Stripe machine payments, and Verifiable Intent & Agent Pay — and how do they differ?
They are the other three payment rails this site covers, alongside x402 — the difference is how each one authorizes and moves money. AP2 & UCP center on mandates: signed, standing permission for what an agent may buy. Stripe machine payments issue agent-scoped payment credentials on Stripe's rails. Verifiable Intent & Agent Pay add provable buyer authorization to card-network payments. The side-by-side view, with citations: the comparison matrix.
Which protocol is going to win?
Nobody knows, and we don't pick sides. The rails are at different stages, built around different tradeoffs. So we enforce neutrality structurally: an identical page template for every rail, one fixed alphabetical order on every surface, and no paid placement. The policy, in writing: our neutrality policy.
Do rules like taxes and consumer law apply to agent purchases?
Yes — using an agent exempts no one from the law. A purchase made by your agent is still a purchase, so tax, consumer-protection, and money-movement rules apply where they normally would. Exactly how they apply is jurisdiction-specific and still being worked out. This is general information, not legal advice.
Can I make money selling to agents?
Honestly: some services earn real, publicly visible revenue from agents, and many listings have never been paid at all. Both halves of that sentence matter. The sellers who get paid can answer a buyer's hardest question — why would I buy this from another agent at all? — with something the buyer's own agents can't do, or can't do well enough. If you sell that, priced per use, the rails to charge for it exist today. When you want your answer on the record, registration is launching too: $5 a month for a published specification and a continuing audit — you pay to be examined, never for an outcome, and unregistered sellers stay just as findable. Practical, step-by-step versions: read the guides.
What does “verified revenue” mean on this site?
Verified revenue is revenue confirmed from public chain data, not self-reported — and attributed per seller address, not per listing. We never take a seller's word for its earnings. One address can sit behind several listings, so sibling listings share one attribution. The precise rule — thresholds and time windows — is defined in the methodology and will be quoted from the engine's spec once frozen.
Why don't you publish the numbers everyone else publishes?
Because we can't check them — and neither can you. Our bar is simple: every number we publish must be recomputable from public data, with a source and an as-of date. Self-reported revenue, listing counts dressed up as market size, and engagement stats all fail that bar. A sentence without a number beats a number without a receipt. The bar itself: our methodology.
Who runs agenteconomy.cloud, and how does it make money?
An independent operation — not owned or paid by any protocol backer, payment company, or seller covered here. Reading is free, with no ads, no affiliate links, and no paid placement. The revenue plan is disclosed plainly: paid tools are launching — buying agents will pay per call to verify sellers before spending, and sellers will be able to pay a monthly fee to be examined — never for a rank, a tier, or coverage. More on About.
Do you accept payment for listings, placement, or coverage?
No. Never. There is no paid placement, no sponsored listing, no affiliate link, and no partnership that buys coverage. If that ever changed, we would announce it visibly — never quietly. The standing policy: neutrality and no paid placement.
How current is this information?
Every page carries a visible “Last reviewed” date, and it is never newer than the page's actual last review. Data sources each get their own as-of dates, published in the methodology's data-freshness section. If something looks stale or wrong, tell us — the contact path is on About.