The Agent API Trial Economy: 10,000 Trials, 323 Services, and What Agents Actually Use
Three weeks ago, minia2a opened its marketplace with a simple premise: agents should be able to discover, try, and pay for API services without human involvement. Today, the platform has processed 10,684 free trials across 323 services from 319 unique users. Here's what the data tells us about the emerging machine-to-machine economy.
The Trial Growth Curve
Agent developers are trying services at an accelerating rate. The platform went from 0 to 10,684 trials in approximately 21 days — an average of ~509 trials per day. But the growth isn't linear. Trial volume has accelerated as more services are listed and more agents discover the marketplace.
At the current trajectory, daily trial volume is roughly 3× what it was in week one. Agents are not just trying more services — they're coming back to try different categories as they discover new capabilities.
The Power Law of Agent Demand
The single most important finding: agent API demand follows a steep power-law distribution. The top 10 endpoints capture 54% of all trials — 5,750 out of 10,684 attempts. Meanwhile, 181 endpoints (56% of the catalog) have fewer than 20 trials each.
| Endpoint | Trials | Unique Users | Category |
|---|---|---|---|
| x402-recall | 1,476 | 72 | Memory / context retrieval |
| x402-captcha-solve | 1,128 | 133 | Web navigation |
| x402-find | 958 | 78 | Search / discovery |
| x402-gas | 861 | 147 | Blockchain utilities |
| x402-polymarket | 445 | 49 | Prediction markets |
| x402-time | 331 | 39 | Utilities |
| x402-web-scrape | 306 | 61 | Web data |
| x402-store | 298 | 33 | Data storage |
| x402-screenshot | 223 | 17 | Web rendering |
| x402-uuid | 193 | 31 | Utilities |
Data source: minia2a.uk/api/stats, August 8, 2026. 323 services total.
The #1 endpoint — x402-recall (1,476 trials) — is a memory/context retrieval service. Agents need to remember things across sessions. The #2 — CAPTCHA solving (1,128 trials, 133 unique users) — reveals that agents spend significant effort navigating human web infrastructure. These aren't exotic AI-to-AI interactions; they're agents doing the unglamorous work of existing in a human-built web.
Category Analysis: What Agents Need
Grouping the 323 services by category reveals clear demand clusters:
| Category | Total Trials | Services | Avg Trials/Service |
|---|---|---|---|
| AI (44 endpoints) | 281 | 44 | 6.4 |
| Web scraping + search (4 endpoints) | 415 | 4 | 103.8 |
| Blockchain gas + tokens (5 endpoints) | 1,058 | 5 | 211.6 |
| Utilities (time, uuid, hash, etc.) | ~800 | ~15 | ~53 |
| Data storage & retrieval | ~350 | ~5 | ~70 |
The most striking pattern: AI services have the most options (44 endpoints) but the lowest per-service trial rate (6.4). Web scraping and blockchain utilities have far fewer options but dramatically higher per-service demand. This is a classic discovery problem: when a category has too many options, agents struggle to find the right one.
Categories with 1–3 services (captcha, recall, gas, polymarket) see concentrated demand — 800–1,500 trials each. Categories with 10+ services (AI, text processing, email) see fragmented demand across many endpoints. The implication for service providers is clear: launch in a thin category, not a crowded one.
The Activation Gap: From Trial to Transaction
Not everything in the data is encouraging. Here's the conversion funnel:
| Stage | Count | Conversion |
|---|---|---|
| Trials | 10,684 | — |
| Unique trial users | 319 | 3.0% of trials → unique user |
| Wallet creators | 55 | 17.2% of users create wallets |
| Paying users | 14 | 25.5% of wallet creators transact |
| Total transactions | 14 | $12.75 total volume |
319 developers tried the API. 55 created wallets. 14 made purchases. This 4.4% trial-to-purchase rate might look low, but it's consistent with developer-tool conversion benchmarks. The bigger question is the wallet-to-transaction gap: 41 developers (75% of wallet creators) have wallets with zero transactions.
Why? Two hypotheses:
- Crypto friction: Funding a wallet with USDC on Base requires crypto-native behavior. Many agent developers come from AI/ML backgrounds and haven't used on-chain payments.
- Free trial sufficiency: With 15 free trials per endpoint, agents may not need to pay yet. The trial economy is rich enough that payment is optional for exploration.
If hypothesis #2 is correct, this is actually good news — it means the paying-user base will grow as agents exhaust their free trials and integrate services into production workflows. The 2,984 credits purchased but only 751 spent supports this: agents are pre-funding for future use.
Credits: A Bridge Between Trial and Payment
The credit system tells an interesting story:
| Metric | Value |
|---|---|
| Credits purchased | 2,984 |
| Credits spent | 751 (25.2%) |
| Credits remaining | 32,133 (includes issued) |
| Credits issued (registration bonuses) | 27,500 |
Credits are being purchased 4× faster than they're spent. This suggests agent developers see enough value to pre-fund accounts but haven't yet reached the sustained-usage phase. The system is accumulating demand-side liquidity — the question is when it converts to spending velocity.
What This Means for the M2M Economy
Three conclusions from 10,684 trials:
1. Agents need web infrastructure, not exotic AI
The top services aren't LLM-based. They're CAPTCHA solving, gas estimation, web scraping, and data retrieval. The M2M economy starts with machines navigating human infrastructure, not machines talking to machines. The bridge services — the ones that let AI agents interact with the pre-AI web — are the highest-demand category.
2. Discovery is the bottleneck, not payment
The x402 payment protocol works. USDC settlement takes seconds on Base. The problem is that with 323 services (and 35,000+ across the broader ecosystem), agents don't know what's available. The 181 services with fewer than 20 trials aren't worse services — they're undiscovered services. The marketplace that solves discovery wins.
3. The trial-to-payment pipeline needs optimization
319 users → 55 wallets → 14 transactions. Each step in the funnel loses ~75-83% of potential users. The biggest drop is user-to-wallet (83% loss). Solving this — likely through fiat on-ramps or wallet abstraction — is the highest-leverage growth lever for any agent payment platform.
The Road Ahead
10,000 trials is a milestone, but it's a mile marker, not a destination. The data tells us that agents want to use paid APIs — they're trying them extensively. The infrastructure for payment works. What's missing is the connective tissue: better discovery, lower payment friction, and trust signals that help agents choose between 323 services.
The M2M economy is real. It's just young. And at 509 trials per day and growing, the data is only going to get more interesting from here.
All data from minia2a.uk/api/stats, retrieved August 8, 2026. minia2a is an open marketplace for agent-to-agent API services using x402 protocol and USDC payments on Base.