Or: why the most-quoted x402 number is measuring the opposite of what most people think it is.

When Coinbase launched Agentic.market in April 2026, the headline stat from the announcement was sharp:

480,000+ transacting agents. 165M+ transactions. ~$50M+ in volume.

That number got picked up everywhere. X threads, news writeups, hot takes about agent commerce. The implication a lot of people drew was that Agentic.market is huge. A catalog of 480K endpoints to choose from.

It is not.

I crawled the actual Agentic.market /v1/services API today, paginated, so I got the whole catalog and not just the first page that most people stop on. Here's what's actually there:

573 services. Not 480K. Not even close.
~29,000 endpoints across those 573 services.
~24,000 distinct URLs after dedup (some endpoints are listed by multiple wrapper providers).
99% of the endpoints settle on Base mainnet. Solana has 90 endpoints. Long-tail chains have a couple each.

The 480K and 29K are different orders of magnitude. So what's the 480K actually counting?

It's the buyers, not the sellers

The Coinbase number is transacting agents. Wallets that have moved x402 USDC across the protocol. That's a demand-side count: every AI agent, every developer-test wallet, every script that's ever paid for an x402 service.

The 28K endpoints in the Agentic.market catalog is the supply-side count. Every distinct paid HTTP endpoint Coinbase's facilitator has observed.

That asymmetry is the actual story.

Flow diagram: ~480,000 transacting agents → 165M+ x402 transactions → ~29,000 endpoints in the Agentic.market catalog

If the demand stat is right, every endpoint in the catalog is being paid by ~17 different agent wallets on average. Most of that is concentrated at the top. A few popular endpoints with thousands of buyer wallets, a long tail with one or two.

Why this matters

The framing changes which problem is the interesting one.

If the catalog is 480K, the interesting problem is discovery. How does anyone find anything in a directory that big? You build search, ranking, recommendation. That's a reasonable problem.

But the catalog isn't 480K. It's 28K. The actual interesting problem is ownership.

Right now, those 28K endpoints are scattered across whoever decided to spin up an x402 endpoint. First-party developers, third-party wrappers proxying brand APIs, paysponge-style aggregators that proxy 50+ distinct services through one infrastructure layer. There's no canonical owner per endpoint, no claim mechanism, no way for the brand whose API is being resold to even know.

And on the buyer side, there are 480K wallets paying for something, but they have no aggregate way to:

Discover which of those 28K endpoints are first-party vs. wrappers
Hold a tradable position in an endpoint they think will grow
Know whether the endpoint they're paying is going to be there in 90 days

Those are coordination problems, not search problems.

What an ownership layer looks like

Full disclosure: I'm building AUTX, which adds an ownership and trading layer on top of Coinbase's discovery layer. The thesis is that Agentic.market built the discovery part well, and the trading part is a separate seat in the same stack.

Here's the rough shape:

Crawl Coinbase's catalog. We hit /v1/services every 6 hours and pull the whole thing (paginated, see footnote on what we initially missed).
Filter to Base mainnet, 1P-only, clean trademark. That's about 600 endpoints today. The rest land in an admin review queue.
Reserve a ticker per endpoint. Pre-claim, the listing is pure discovery. POST /proxy/<ticker> returns a 302 redirect to the source endpoint. AUTX takes zero revenue.
Wait for the original developer to claim. Free, no gas. They sign from the merchant wallet (EOA signature) or prove domain control. After claim, they can deploy a per-agent ERC-20 token on a bonding curve (~$0.20 gas) and start trading.
Post-claim economics: AUTX charges a 10% platform fee. Details below.

AUTX isn't the only or best answer to this. The point is that "ownership of an x402 endpoint" is a missing primitive in the current stack, and somebody is going to build it. If the demand side really is 480K wallets, the floor for what an ownership layer is worth is not small.

What the developer actually keeps

There's persistent confusion in the comments about AUTX charging "28%" because people see the 10/72/18 split and add up the wrong numbers. So let's show the actual flow.

When a buyer pays $1.00 for an x402 service through AUTX, here's what happens on-chain, every order:

Flow diagram: $1.00 buyer payment splits on-chain — $0.72 direct USDC to dev wallet, $0.18 buyback-and-burn on the dev's own agent token, $0.10 platform fee to AUTX

Out of every dollar:

72¢ goes direct to the dev's wallet. Settled in USDC on Base L2, every order, no escrow, no holds.
18¢ buys back and burns the dev's own agent token. That's not a fee. It's mechanical supply compression: fewer tokens in circulation against the same demand. The dev and their token holders own the upside of that compression. The buyback executes via on-chain transaction, verifiable on BaseScan.
10¢ is the AUTX platform fee. Goes to the AUTX DAO treasury.

The math people actually want: AUTX takes 10%. The other 18% is value that stays in the dev's economy.

For comparison:

Virtuals Protocol (tokenless ACP): 5-10% flat fee. AUTX's 10% is in the same ballpark, with a token-economics layer underneath instead of a flat take.
App Store / Google Play: 15-30% take rate.
RapidAPI: 25% take rate, no token economics.
GPT Store: Pays $0 to developers.

If you're building an agent that earns real revenue, the 10% platform fee is competitive with anywhere else you'd list it. The 18% buyback-burn is upside on top, not a cost.

The footnote on what I initially missed

When I first pulled Agentic.market's catalog, I got 50 services / 1,200 endpoints. That's 8.7% of the actual catalog.

The bug: I was hitting GET /v1/services without ?limit= or ?offset= params. The API defaults to 50 services per page and returns a total field in the response. Without pagination, you only ever see the first page.

I'm not the only one. Half the X posts about Agentic.market's catalog size are using that 50-service number, or the 1,200-1,800 endpoint count that comes from it. The real number is bigger by an order of magnitude. The demand/supply asymmetry is sharper than the first-page snapshot suggests.

If you're crawling Agentic.market for any reason, paginate. Use:

curl -H "User-Agent: …" \
  "https://api.agentic.market/v1/services?limit=200&offset=0"

and loop with offset += len(services) until offset >= total. Server caps page size at 200.

Where this lands

Three things to take away:

The 480K number is real. It just doesn't mean what most posts imply. It's demand-side wallet count, protocol-wide. The catalog of things you can pay is 28K, not 480K.

The seller-side concentration is the interesting market. About 600 endpoints today are first-party-vetted, base-mainnet, brand-clean. That's the universe where ownership transfer and trading economics actually have something to land on.

If you're sizing any market opportunity off Agentic.market data, paginate the API or you're sizing off 8% of the actual catalog.

Lance Lukens is building [AUTX](https://autx.ai), a marketplace and bonding-curve trading layer for AI agents that earn real revenue. AUTX shadow-lists every 1P x402 endpoint on Base mainnet from Coinbase's Agentic.market catalog, with a free claim path for the original developer. Post-claim, devs keep 72% of every order in USDC. AUTX charges a 10% platform fee and the remaining 18% buys back and burns the dev's own agent token. Wyoming DAO LLC, March 2026.