What 12,590 Agent API Calls Reveal About the M2M Economy
The Top 10: Agents Need Infrastructure, Not AI
When you look at what agents actually call — not what sounds impressive in a pitch deck — the pattern is unmistakable. The top 10 most-called endpoints on minia2a are not AI models, not trading signals, not "agent skills." They're infrastructure utilities.
| # | Endpoint | Trials | Users | Category |
|---|---|---|---|---|
| 1 | x402-recall — agent memory | 1,768 | 94 | Storage |
| 2 | x402-gas — gas price lookup | 1,248 | 172 | Blockchain |
| 3 | x402-captcha-solve — CAPTCHA solving | 1,202 | 133 | Web |
| 4 | x402-find — web search | 1,144 | 83 | Search |
| 5 | x402-polymarket — prediction markets | 624 | 63 | Finance |
| 6 | x402-time — current time | 531 | 55 | Utility |
| 7 | x402-screenshot — page screenshots | 306 | 20 | Web |
| 8 | x402-web-scrape — page scraping | 306 | 61 | Web |
| 9 | x402-store — key-value storage | 298 | 33 | Storage |
| 10 | x402-uuid — UUID generation | 225 | 33 | Utility |
Let that sink in. The #1 thing agents pay for is memory. The #3 thing is CAPTCHA solving. UUID generation — a one-liner in any language — has 225 calls. These are not exotic AI capabilities. They're the digital equivalent of a hammer, a screwdriver, and a measuring tape.
Gas Prices: The Killer App Nobody Talks About
x402-gas has 172 unique users — the most of any endpoint. That's 2.8× more unique users than the #1 endpoint by volume. This tells us something profound: agents that interact with blockchains need to know gas prices constantly. It's not a one-time lookup. It's a recurring need at decision time.
This makes sense when you think about it. An agent deciding whether to execute a trade, bridge assets, or deploy a contract needs current gas data. Getting it wrong costs real money. Getting it right requires a reliable, always-available data source — exactly the use case for a paid API.
The gas endpoint's user-to-trial ratio (172 users / 1,248 calls = 7.3 calls/user) is lower than recall's (94 users / 1,768 calls = 18.8 calls/user), suggesting gas is used more broadly but less intensively — many agents need it occasionally. Recall, by contrast, has fewer users making more calls — the pattern of a core dependency.
The Long Tail: 60% of Services Are Invisible Without Free Trials
181 of 328 endpoints (55%) have fewer than 20 trials. Without free trials, these services would be practically invisible to agents. Discovery in the M2M economy is not a solved problem — it's the bottleneck.
This is not a failure — it's the expected shape of any marketplace. But it means the discovery mechanism is the marketplace. If agents can't find a service, it doesn't matter how good it is. Free trials are the bridge between "exists" and "discovered."
The endpoints that do break through tend to be named intuitively: gas, time, uuid, captcha-solve. The naming is the interface. An agent that needs a gas price will search for "gas." An agent that hits a CAPTCHA will search for "captcha." Self-describing names are not a nice-to-have — they're the primary discovery mechanism in an agent-driven economy.
The Conversion Gap: 62 Wallets, 14 Transactions
Here's the uncomfortable number: 62 agents have registered wallets on minia2a. Only 14 have ever made a paid transaction. That's a 22.6% wallet-to-paid conversion rate.
But the deeper number is more interesting: 3,096 credits have been purchased, but only 816 have been spent. People are putting money into the system and then walking away. The payment infrastructure works. The trial experience works. What's broken is the path from "I have credits" to "I am spending them regularly."
This pattern — buy, then abandon — is familiar from SaaS free trials. But the fix is different for agent payments. Agents don't browse. They don't get "onboarding emails." They integrate once and then call programmatically. If that integration doesn't happen in the first session, it probably never happens.
The implication for API builders: your documentation is your onboarding. The curl example on your service page is not decoration — it's the difference between a registered wallet and a paying customer. Every endpoint needs a copy-paste-ready example, a clear price, and a reason to call it a second time.
What This Means for the M2M Economy
1. Infrastructure utilities are the beachhead
The first wave of M2M payments is not for AI inference. It's not for "agent skills." It's for the boring, essential utilities that agents need to function: memory, time, search, CAPTCHAs, gas prices. These are the APIs that will drive the first billion agent transactions — not because they're exciting, but because they're necessary.
2. Discovery is the unbundled marketplace
In the web economy, Google solved discovery. In the app economy, app stores solved discovery. In the M2M economy, discovery is fragmented across protocol indexes, marketplace directories, MCP registries, and agent frameworks. The marketplace that solves "how does my agent find the right API at runtime?" wins.
3. Free trials are not a growth hack — they're the product
When 55% of services have fewer than 20 trials, the trial mechanism isn't a marketing feature. It's the only way the long tail exists at all. Without free trials, the M2M economy collapses to the top 10 endpoints — and those top 10 are all commodity utilities. The value is in the long tail, and the long tail only works with frictionless discovery.
4. The naming is the interface
Agents don't read marketing pages. They search for function names. gas beats blockchain-fee-estimator-v2 every time. The most successful endpoints on minia2a have short, self-describing names. This is a design principle that most API builders haven't internalized yet.
5. "Buy then abandon" is the conversion problem to solve
3,096 credits purchased, 816 spent. This is not a payment problem — the payment works. It's an integration problem. The window between "I bought credits" and "I integrated the API" is where conversion lives or dies. Every minute that passes without a successful API call reduces the probability of a second one.
What API Builders Should Do
- Name it what it does. No branding, no version numbers, no cleverness.
gas,time,captcha-solve. - Make the first call zero-friction. Free trial, no API key, no registration. The agent needs to get a result in one HTTP request.
- Document with a copy-paste curl command. Not a paragraph. Not a tutorial. A single curl command that returns a useful result.
- Price per call, not per month. Agents don't have subscription budgets. They have per-task budgets. $0.01/call beats $10/month for agent customers.
- Be reliable enough to depend on. Recall has 1,768 calls from 94 users (18.8 calls/user) because it works every time. Reliability = repeat calls.
The Bottom Line
The M2M economy is real — 12,590 trial calls, 438,000+ total requests, real USDC changing hands. But it doesn't look like what most people imagine. It's not agents hiring other agents for complex tasks. It's not autonomous negotiation between AI systems. It's agents making simple, repeatable utility calls — get the gas price, solve this CAPTCHA, remember this fact, search for that thing.
The companies that win in this space won't be the ones with the most impressive AI. They'll be the ones that make it boring to pay for an API call. Infrastructure that works so reliably you forget it's there. Discovery that surfaces the right endpoint without thinking. Integration that takes one curl command.
The M2M economy doesn't need more magic. It needs better plumbing.