The x402 protocol now processes over 200 million transactions. Agentic.market — Coinbase's own marketplace — lists 2,056 services. OKX launched an AI marketplace with escrow contracts. The discovery layer is being built by multiple players simultaneously.
But there's a problem hiding in plain sight. And it's not about how many services are listed. It's about how agents choose which ones to call.
Across minia2a's 323 services and 11,196 free trial calls, the distribution is stark:
| Trial Range | Endpoints | % of Total |
|---|---|---|
| 1,000+ trials | 4 | 1.2% |
| 100–999 trials | 12 | 3.7% |
| 20–99 trials | 27 | 8.4% |
| Fewer than 20 trials | 181 | 56.0% |
| Fewer than 10 trials | 99 | 30.7% |
Just 10 endpoints — 3% of the catalog — capture 55% of all trial activity. Recursive memory retrieval (x402-recall), CAPTCHA solving, gas estimation, and semantic search (x402-find) dominate. These are the endpoints agents already know they need.
Meanwhile, 181 endpoints have fewer than 20 trials. 99 have fewer than 10. These aren't bad services — they're undiscovered ones. A color palette generator, a CSV-to-JSON converter, a chain registry lookup, a QR code renderer. Utilities that agents would use if they knew they existed.
Raw trial count tells you what's popular. But stickiness — unique users per trial — tells you what's actually useful once discovered:
| Endpoint | Trials | Unique Users | Stickiness (users/trial) |
|---|---|---|---|
| x402-color-palette | 20 | 13 | 0.65 |
| x402-csv-json | 20 | 12 | 0.60 |
| x402-chain-registry | 22 | 13 | 0.59 |
| x402-hash | 40 | 23 | 0.57 |
| x402-country | 21 | 12 | 0.57 |
Compare this to the high-volume endpoints:
| Endpoint | Trials | Unique Users | Stickiness |
|---|---|---|---|
| x402-captcha-solve | 1,177 | 133 | 0.11 |
| x402-gas | 1,013 | 152 | 0.15 |
| x402-recall | 1,556 | 78 | 0.05 |
The pattern is clear: niche utility endpoints convert browsers into regular users at 4–6× the rate of the popular gateway endpoints. A CAPTCHA solver gets 1,177 trials from 133 users — most try it once. A color palette generator gets 13 users from 20 trials — almost everyone who finds it keeps using it.
Discovery without trials kills the long tail. If agents had to pay to try x402-color-palette, it would have 0 trials, 0 users, and 0 chance of ever being discovered. The 13 agents who now use it regularly would never have known it existed.
As of August 2026, the x402 marketplace landscape is splitting into two models:
Both models have a place. Pay-first works for known quantities — agents that need Bloomberg data or Google Flights already know what they're buying. Trial-first works for discovery — agents exploring 35,000+ possible tools (across Coinbase's 14,865 and Apify's 20,000) need to try before they commit.
minia2a's funnel from trial to payment reveals exactly where the friction lives:
11,196 trials → 319 unique users → 56 wallets → 14 paid transactions ($12.75) 100% 2.85% 0.50% 0.125%
Two-thirds of wallet creators (42 out of 56) have never made a paid transaction. The protocol works — settlement is instant on Base, fees are under $0.01. The bottleneck isn't payment infrastructure. It's integration friction and the discover-to-trust pipeline.
This is exactly what you'd expect at this stage. Agents try things for free. The ones that find something useful come back. Some of those eventually pay. The funnel is wide at the top and narrow at the bottom — because it should be. Every paid transaction represents an agent that went through: discover → try → evaluate → integrate → pay.
The x402 protocol has solved the payment problem. USDC settlement on Base is essentially frictionless — 2-second confirmations, sub-cent fees, no chargebacks. Cloudflare Wallets, Stripe's crypto APIs, and Mastercard's AP4M are all building on top of this.
The next problem isn't "can agents pay?" — it's "how do agents find the right thing to pay for among 35,000 options?"
Free trials are the answer. Not as a growth hack or a marketing tactic, but as the only mechanism that makes the long tail economically viable. Without trials:
With trials, the long tail breathes. A QR code generator gets 17 regular users. A domain intelligence tool gets 10. A JWT decoder gets 19. Each one serves a specific agent need that would otherwise go unmet — or would be filled by a more expensive, less specialized alternative.
The agent API economy is at an inflection point. Payment infrastructure is commoditizing. Discovery layers are being built. The marketplaces that win won't be the ones with the most services — they'll be the ones that let agents find the services they actually need.
Free trials aren't a nice-to-have. They're the discovery mechanism for the long tail. And in an economy where 56% of services get fewer than 20 discovery attempts, the long tail is where most of the value lives.
Data source: All statistics from the minia2a /api/stats endpoint, queried August 8, 2026. 11,196 total trials across 323 services. Competitive data from public documentation and marketplace listings for Agentic.market and OKX AI Marketplace.
Correction/update? Open an issue or contact [email protected].