We categorized all 252 trial-enabled endpoints and measured actual usage — not signups, not pageviews, but real API calls made by agents during trials. The pattern is clear:
Crypto/DeFi and AI/ML are effectively tied for first place (24.1% vs 23.9%). Together they represent nearly half of all agent trial activity. This matters because it tells us what agents are already doing — before auto mode, before autonomous spending, before the ecosystem matures. These are the use cases agents reach for first.
The top two categories — Crypto/DeFi and AI/ML — together account for 48% of all trials. If you're building an agent payment API, the demand is here.
Revealed preference — what agents actually call, not what they say they want. Here are the top 10 endpoints by trial volume:
| Endpoint | Category | Trials | Unique Users | Price |
|---|---|---|---|---|
| Mythos x402 Recall | AI/ML | 1,997 | 122 | $0.005 |
| Multi-Chain Gas Price | Crypto/DeFi | 1,685 | 226 | $0.005 |
| Mythos x402 Bazaar Find | AI/ML | 1,294 | 101 | $0.005 |
| CAPTCHA Solver | Security | 1,203 | 133 | $0.001 |
| Time Utilities | Utility | 970 | 102 | $0.001 |
| Polymarket Odds Live | Crypto/DeFi | 897 | 109 | $0.005 |
| URL Screenshot (PNG) | Web/Data | 381 | 24 | $0.01 |
| UUID v4 Generator | Utility | 348 | 41 | $0.001 |
| Web Scraper | Web/Data | 310 | 64 | $0.005 |
| Mythos x402 Memory Store | AI/ML | 300 | 34 | $0.005 |
Gas Price has the most unique users (226) — every agent that touches a blockchain needs to know current gas. Recall and Find are the AI-native discovery endpoints agents use to locate other services. CAPTCHA solving (133 users, $0.001/call) is the cheapest paid endpoint and sees heavy use, suggesting agents routinely encounter CAPTCHA walls and need programmatic solutions.
752 registered agents hold 379,900 credits. Only 842 have been spent — a 0.22% utilization rate. This looks alarming until you understand the context:
Agents can't spend autonomously yet. Today, a registered agent needs a human operator to manually authorize each paid API call. The 14 paid transactions on the platform all came from developers who manually triggered payments — not from autonomous agents making spending decisions.
Claude Code auto mode (Aug 14) changes this. When an agent has permission to execute tool calls without human approval, and those tool calls include x402 payment endpoints, the agent can autonomously decide: "I need the current gas price → that endpoint costs $0.005 → it's within my $5/day budget → pay and proceed."
The 0.22% credit utilization rate isn't a product failure. It's a pre-auto-mode baseline. Aug 14 is when we find out whether agents, given the ability to spend autonomously, actually will.
Three things unlock simultaneously:
{daily_limit_usdc: 5, max_per_call_usdc: 1}) gives agents a hard spending cap. Combined with machine-readable 402 headers (x-402-amount, x-402-chain, x-402-token), an agent can check price against budget before paying — autonomously, safely, verifiably.15,136 trials are not random noise. They are the revealed preference of agents exploring what's available. The pattern — Crypto/DeFi + AI/ML dominating, utility and security as essential infrastructure, finance as a small but high-value niche — tells API builders exactly where to focus.
For the 752 registered agents sitting on 379,900 unused credits: Aug 14 is when the switch flips. The infrastructure is in place. The safety case is made. The budget caps are set. All that's left is for agents to start spending.