In the last week, agents have made 5,769 free trial calls across 160 different x402 API endpoints on minia2a. They've tested captcha solvers, gas estimators, wallet scanners, token security checkers, Polymarket data feeds, and a hundred other microservices. The trial economy is real — agents are actively exploring what's available.
But when you follow the funnel from trial to paid, the numbers tell a different story.
14 paid transactions. $12.75 in total volume. 0.24% conversion from trial to paid.
This isn't a failure of the protocol. x402 works. The payment rails are live on Base, Algorand, Casper, XDC, BNB Chain, and Bitcoin Lightning. Visa, Mastercard, Stripe, and AWS have all backed the standard through the Linux Foundation. The infrastructure is ready.
The data reveals something deeper: agents want to consume APIs. They just can't pay for them — yet.
The trial distribution is revealing. When you sort endpoints by trial volume, the top looks like a utility belt for automation scripts:
| Endpoint | Trials | Registered Users | Category |
|---|---|---|---|
| x402-captcha-solve | 662 | 0 | Web automation |
| x402-gas | 430 | 0 | Blockchain utility |
| x402-find | 370 | 0 | Search/text |
| x402-recall | 329 | 0 | Memory/knowledge |
| x402-web-scrape | 304 | 25 | Web data |
| x402-store | 275 | 0 | Storage |
| x402-time | 207 | 15 | Utility |
| x402-polymarket | 156 | 0 | Prediction markets |
| x402-token-security | 136 | 19 | DeFi security |
| x402-fear-greed | 111 | 0 | Market sentiment |
The pattern is striking: the highest-volume trial endpoints have zero registered users. Captcha-solve (662), gas (430), find (370), recall (329), store (275) — combined, these five endpoints account for 2,066 trials (36% of all trial activity) but not a single one of their callers has created an account.
These look like automated scripts and bots testing the free tier — not agents with wallets preparing to pay. They're treating the trial endpoints as free utilities, not evaluating a paid service.
When you sort by registered users instead of trial volume, a completely different picture emerges:
| Endpoint | Users | Trials | Category |
|---|---|---|---|
| x402-web-scrape | 25 | 304 | Web data extraction |
| x402-token-security | 19 | 136 | DeFi security audit |
| x402-time | 15 | 207 | Date/time utility |
| x402-uuid | 13 | 100 | ID generation |
| x402-wallet | 9 | 58 | Wallet operations |
| x402-wallet-intel | 8 | 45 | On-chain intelligence |
| x402-base64 | 7 | 47 | Encoding utility |
These are developer tools and data APIs — the kind of endpoints you'd integrate into an agent's tool chain. Web scraping, token security, time utilities, UUID generation, wallet operations. These callers are evaluating the service for real integration, not just grabbing free compute.
27 endpoints have more than 10 trials but zero registered users. Only 52 out of 160 trial-enabled endpoints (32.5%) have attracted any registered users at all. The "try it for free" funnel is wide at the top and extremely narrow in the middle.
The 0.24% trial-to-paid conversion rate isn't just a minia2a problem — it reflects where the entire M2M payment ecosystem is right now. The payment rails exist. What's missing is everything around them:
Most AI agents today are built without a payment module. They can call free APIs. They can use trial credits. But when the credits run out, they stop — because nobody gave them a USDC balance and a spending policy. The agentwallet-sdk and x402 payment protocol exist, but they haven't been integrated into the major agent frameworks (CrewAI, AutoGPT, LangChain) as a default capability.
When an agent exhausts its 500 free credits on minia2a, the next step requires a human to add USDC to a wallet. That's the moment the funnel breaks: the agent can't top up its own wallet. The 35 wallets that exist (0.61% of trial users) represent humans who completed this manual step — a step that shouldn't exist in a machine-to-machine economy.
An agent that finds a useful endpoint on its first trial call has no built-in mechanism to remember it and return. There's no agent-native service registry, no automatic discovery, no "recently used services" feed. Each trial is a one-off. The 96.5% of x402 listings with zero repeat customers (documented in our discovery crisis analysis) reflects this systemic gap.
While agent spending behavior lags, the infrastructure layer is accelerating:
The payment rails are being built at an extraordinary pace. Five chains shipped x402 facilitators in the last 72 hours. But the gap between "can pay" and "does pay" is where the real work remains.
This is the uncomfortable truth in the data: we've built a checkout counter that works, but nobody taught the customers how to shop. The 5,769 trials prove that agents want these APIs. The 14 transactions prove that the payment rails function. The chasm between them — a 99.76% drop-off — proves that the missing piece isn't technology.
What needs to happen next:
1. Agent frameworks need payment modules by default. LangChain, CrewAI, and AutoGPT agents should ship with the ability to hold and spend USDC. This should be as standard as an HTTP client.
2. Automatic wallet provisioning at registration. The current flow — "sign up, then manually add USDC" — kills conversion. Agents should get a wallet with credits the moment they register.
3. Agent-native discovery. Agents need a machine-readable service catalog they can query programmatically — not a human-facing marketplace. minia2a's /api/services endpoint is a step, but discovery needs to be built into the agent's runtime.
4. Spending policies, not just wallets. Developers won't give agents open-ended spending authority. The Casper guardrail model — programmable constraints per agent — needs to become the standard.
We're not just observing this gap — we're building the bridge:
/api/services — machine-readable endpoint catalog for agent runtime integration.The infrastructure is ready. The trials prove demand. The 14 transactions prove the rails work. Now we need to connect the two — and that's the work of the next 12 months for everyone building in the M2M payment space.