Across 299 x402 services and 7,504 agent API trials, a brutal power law is at work. Ten endpoints — 3.3% of the catalog — capture 54% of all agent trial activity. Twenty endpoints capture nearly two-thirds. At the other end, 181 services have fewer than 20 trials each.
This isn't surprising to anyone who's spent time in marketplaces. But the numbers are stark — and they have real implications for anyone building agent-facing APIs.
Here are the ten endpoints that dominate agent trial activity, ranked by total trials:
| # | Endpoint | Trials | Users | Trials/User | Category |
|---|---|---|---|---|---|
| 1 | x402-captcha-solve | 914 | 133 | 6.9 | Web Navigation |
| 2 | x402-recall | 656 | 43 | 15.3 | Memory |
| 3 | x402-find | 558 | 61 | 9.1 | Search |
| 4 | x402-gas | 536 | 119 | 4.5 | Blockchain |
| 5 | x402-web-scrape | 306 | 61 | 5.0 | Web Navigation |
| 6 | x402-store | 291 | 32 | 9.1 | Memory |
| 7 | x402-time | 282 | 34 | 8.3 | Utility |
| 8 | x402-polymarket | 215 | 31 | 6.9 | Prediction Markets |
| 9 | x402-token-security | 150 | 40 | 3.8 | Blockchain |
| 10 | x402-uuid | 147 | 26 | 5.7 | Dev Tools |
| Top 10 Total | 4,055 | — | — | 54.0% of all trials |
When you plot trials per endpoint from most-used to least-used, the curve is unmistakable:
| Tier | Endpoints | % of Catalog | Total Trials | % of Trials |
|---|---|---|---|---|
| 🏆 Top 10 (100+ trials each) | 11 | 3.7% | 4,111 | 54.8% |
| 🥈 Tier 2 (50–99 trials) | 10 | 3.3% | 731 | 9.7% |
| 🥉 Tier 3 (20–49 trials) | 39 | 13.0% | ~1,220 | 16.3% |
| 📉 Long Tail (1–19 trials) | 181 | 60.5% | ~1,442 | 19.2% |
| ❌ Zero trials | 58 | 19.4% | 0 | 0% |
60% of the catalog shares less than 20% of the trial volume. Meanwhile, 11 endpoints — the ones with 100+ trials — capture more than half of all activity. This is Zipf's Law playing out in the agent API economy.
An AI agent deciding which API to call has limited context. It can't browse 299 services the way a human scrolls through an app store. It sees whatever its developer configured — usually the first 5–10 tools that match a keyword. The result: popular endpoints get more popular because they're the ones agents find first.
Agent developers share code. A developer sees someone else's agent using x402-captcha-solve, copies the tool definition, and now another agent is using it. No developer is searching for "what's the 47th most popular x402 endpoint" — they use what they see others using.
On most platforms, trying a new API costs money — even if it's a fraction of a cent. An agent with a $5 budget isn't going to experiment with 50 different text-processing endpoints to find the best one. It'll use the first one that works. The power law gets more extreme the higher the cost of experimentation.
This is where minia2a's model changes the game. Every endpoint on the platform offers free trials — no wallet, no payment, no commitment. Agents can test an endpoint, verify it works, and only pay when they need production volume.
Why this matters for the power law:
x402-qr, 27 trials) risk-free. If it's good, they integrate it. The long tail has a chance.
We can see this in the data. 241 out of 299 endpoints have at least one trial — that's 80.6% of the catalog getting some discovery. On paywall-gated platforms, you'd expect the bottom 50% to get zero activity. The fact that 181 endpoints in the long tail have at least some trial activity is evidence that free trials work as a discovery mechanism.
If you're building an agent-facing API, the power law has practical implications:
1. Category choice matters more than quality. The top 10 endpoints aren't necessarily the best at what they do — they're in categories agents need most (web navigation, memory, blockchain). Building the world's best barcode generator won't overcome the fact that agents don't need barcodes very often.
2. Free trials are your only distribution. In a power-law world, you can't out-market the top 10. Your only hope is that agents try your endpoint — and the only way they'll try it is if trying is free. If you're behind a paywall, you don't exist.
3. Stickiness beats popularity. x402-recall has 43 users — 3x fewer than x402-captcha-solve (133). But recall users call it 15.3 times each vs. captcha's 6.9. A niche endpoint with 20 devoted users who call it 50 times each generates more sustainable volume than a broad endpoint with 100 users who call it twice. In the long tail, retention is your growth strategy.
4. The catalog isn't the product — the ranking is. With 299 services, no agent can evaluate them all. The ranking mechanism — whether it's trial count, user ratings, or something else — determines which endpoints get discovered. If the ranking is gameable, the power law gets gamed. If it's based on real usage, the best endpoints rise.
The agent API economy obeys the same power-law dynamics as every other marketplace — app stores, e-commerce, content platforms. Most of the value concentrates in a small number of endpoints. Most endpoints get very little attention.
But there's a difference: agent marketplaces are still small enough that the power law hasn't hardened. At 7,504 trials, the #1 endpoint has 914 trials — not 914,000. The rankings are still fluid. A new endpoint that solves a real agent problem can still break into the top 20 within weeks.
That window won't stay open forever. As the M2M economy scales from thousands to millions of agent transactions, today's power-law distribution becomes tomorrow's locked-in hierarchy. The endpoints that establish themselves now — while discovery is still open and experimentation is free — will be the ones agents default to when the economy is 100x larger.