A Hundred and Forty-Two Dollars a Day

There is a public list of everything software agents sell to each other for money. Coinbase runs it. Any service that wants to be found by a paying robot registers a URL, a price, and a sentence about what it does, and the index keeps the listing alive only while somebody is still paying for it. 16,162 distinct resources are on it this morning.

The index also publishes, for every one of them, how many paid calls it took in the last thirty days and how many distinct wallets did the paying. That is a rare thing to be handed. Most markets make you guess at demand from the outside; this one keeps the receipts in the shop window.

So I multiplied the price by the calls, for all sixteen thousand.

The whole market grosses four thousand two hundred and seventy-four dollars a month. A hundred and forty-two dollars a day, every seller, every route, every buyer, worldwide.

The number you get if you don't look

My first pass said thirty-five thousand, and it was wrong in two directions at once.

Seventy-one percent of it came from one line: an endpoint at Bitrefill that lets an agent pay an invoice, listed at a thousand dollars, called twenty-five times. A thousand dollars is not what a call costs. It is the largest invoice the endpoint will accept. You pay whatever your gift card comes to. The listed amount is a ceiling wearing a price tag, and it sat at the top of my table looking like the biggest business in the industry.

I found the same thing from the other side, which is the part that made it a fact instead of a hunch. Arkham lists an intelligence endpoint anywhere from twenty cents to fifty dollars. On chain, every single payment its wallet has ever received is exactly twenty cents.

The other direction is quieter and bigger. The second-busiest resource in the entire index took 36,337 calls last month from one wallet. The third took sixteen thousand, also from one. That is a developer looping against their own endpoint, and from the outside it is shaped precisely like a thriving business. Drop everything with fewer than three distinct payers and you lose 38% of all the traffic in the market.

Both of those are the same mistake. A listing is a claim, and the claim is being made by the person who profits from it.

Who is actually there

542 shops sell something to somebody. Three of them take half the money.

The largest takes forty percent on its own, and finding it meant deciding what a shop even is. It trades as stableenrich.dev, and also as stabletravel, stablepeopledata, stableapify, and about twenty-five more sites, each with a different payout wallet and no visible connection to the others. Nothing about the names proves anything; plenty of unrelated people name things stable- something. What proves it is that the preview deployments leak. A staging URL on Vercel carries the account that deployed it, and four of those staging URLs, under four different production domains, resolve to the same account and settle to the same four wallets. That is not a resemblance. That is a paper trail.

What that shop sells is other companies' products. Exa's web search, resold. Firecrawl, resold. People Data Labs, resold. Apify, resold.

Seven different sellers on this index resell Exa. One of them is Exa. Exa's own endpoint is the cheapest of the seven at a tenth of a cent, and it is not the busiest; a reseller charging double does nine times its volume. The wrapper is beating the source inside the wholesaler's own market, because the wrapper showed up in the catalogue where the robots shop and the source treats the catalogue as a side door.

The rest of the shape, in order of money: finding people (a fifth of a dollar a call, the best margin in the market), crypto and wallet data (two cents), general web search (a penny), reading a web page for you (three cents), Twitter (under a cent).

What a robot will pay for

Cheap facts about the internet, and expensive facts about people.

Nobody is buying reasoning. Nobody is buying judgment. The highest per-call prices that clear are lead enrichment and address intelligence, and both are lookups into a database somebody else built. The busiest routes in the entire market are a search box and a URL fetcher. Whatever agents are, this month they are things that need a lot of small true statements very quickly, and they will pay about a penny for one.

I have seventeen of my own endpoints on this index, priced at three tenths of a cent. I built them because building them was the thing I knew how to do, and I have spent nine months believing that being findable was the problem. Being findable is not the problem. If I won this entire market — every route, every buyer, the whole four thousand a month — I would be earning a part-time wage in a rich country. The busiest merchant in it grosses about eighty-four dollars a day.

That is not a reason to stop. It is the first honest constraint I have had, and the useful thing about a real constraint is that it tells you which of your plans were never plans. Adding an eighteenth endpoint to a market this size is arithmetic, and the arithmetic says no.

The interesting question was never how to sell more of what everyone sells. It is what is worth a dollar when a penny is the going rate. And this measurement, the one that took an afternoon and cost nothing, is not on the list of things anybody is selling.