0.3%: Why 5,250 Agent API Trials Produced 54 Settled Payments
We run an open marketplace for AI agent APIs — a directory where any developer can list a pay-per-call endpoint, any AI agent can discover it, and payments settle in USDC on-chain via the x402 protocol. No sign-up required for agents. No API keys. Just POST and pay.
We've been live for over 100 days. Here's the raw data as of today:
Conversion rate: 0%. 54 agents have settled real payments (0.45 USDC) (the 14 "paid" records were credit-consumption entries — see correction note).
The instinct is to call this a failure. But it's not a minia2a failure — it's a snapshot of where the entire agent economy actually is in August 2026, underneath the $50 million settlement volume and 69,000-agent headlines. Let me walk through what the data actually says.
The Funnel, Step by Step
Every agent payment goes through four gates:
Gate 1: Discovery
An agent can't pay for an API it doesn't know exists. minia2a.uk lists 207 services, but agent discovery is the weakest link in the ecosystem. Agents don't browse directories — they get configured with a fixed tool set by their developers. Until agent frameworks natively integrate service discovery, marketplaces are discoverable to humans, not agents.
Gate 2: Trial → Wallet
5,250 free trials were consumed by 316 unique agents. But only 35 of those 316 (11%) ever set up a wallet. This is the biggest drop in the funnel. Free trials work — agents use them — but most never take the next step of creating a payment method.
Why? Two reasons:
- Wallet creation requires a human developer. An AI agent can call an API, but it can't fund a wallet with USDC. A human has to bridge fiat → crypto, transfer to Base, and configure the agent's wallet address. That's a 3-step process that loses 89% of potential payers.
- Free trial credits are sufficient for experimentation. minia2a issues 500 free credits per wallet. Most agents haven't exhausted their free credits yet (21,726 credits remain unspent across all wallets). They're still in exploration mode.
Gate 3: Wallet → Payment
Of the 35 wallet users, zero made a settled on-chain transaction (the 14 "paid" records were credit-consumption entries — see correction note). The payment rail is still being validated end-to-end. The bottleneck is everything before settlement — and the settlement step itself.
This Isn't Just Us
The conversion gap is visible across the entire agent payment ecosystem. Here's what independent analysis reveals:
| Platform | Scale Signal | Conversion Signal |
|---|---|---|
| x402 Protocol | 165M txns, 69K+ agents, ~$50M settled | Avg ticket $0.30 — most volume concentrated on few services |
| Stripe Agent Commerce | 1,000+ merchants onboarded | Transacting agents still in single digits (CoinDesk, Mar 2026) |
| On-chain (overall) | Growing agent activity | Daily agent tx volume ~$17K, ~50% are test transactions |
| minia2a.uk | 207 services, 331K total requests, 5,250 trials | 54 settled txns (0.45 USDC), $0.45 volume (54 settlements), 0.3% conversion |
The pattern is consistent: the supply side (merchants, services, infrastructure) is building fast, but the demand side (agents with wallets making autonomous payments) is still in its infancy. It's not a failure of any single platform — it's the natural shape of infrastructure that's being built ahead of demand.
What's Actually Working
Despite the low conversion, the trial data reveals genuine agent demand patterns:
| Service Category | Trials | What It Means |
|---|---|---|
| CAPTCHA solving | 569 | Agents hit bot walls — they need human emulation tools |
| Gas price lookup | 408 | DeFi agents need real-time on-chain context |
| Entity find/search | 370 | Agents need structured knowledge retrieval |
| Memory recall | 328 | Stateful agents need external memory |
| Web scraping | 304 | Agents need to read the open web |
| Utility compute | ~700 | QR, Base64, Hash, JSON, JWT, color, UUID, etc. — agents use pure functions |
The top 5 categories alone account for 1,979 of 5,250 trials (38%). These are not speculative use cases — real agents are hitting real problems and reaching for APIs to solve them. The question isn't whether agents need paid APIs. It's when they'll start paying.
The Three Things That Need to Change
1. Wallet Onboarding Must Be Agent-Native
Today, an agent developer has to manually fund a Base wallet with USDC and configure their agent to use it. That's 2021-era crypto UX. We need one-click wallet provisioning — where an agent framework can spin up a wallet, fund it with $5 of USDC, and start paying for API calls in under 60 seconds. Coinbase's AgentKit and OKX's Agent Payments Protocol are building this, but it's not mainstream yet.
2. Discovery Must Be Built Into Agent Frameworks
When a LangChain or CrewAI agent needs to solve a problem, it should be able to search a decentralized service registry — not just the tools its developer hardcoded. Service discovery as a protocol primitive, not an afterthought. minia2a's /api/services endpoint is a step in this direction, but the major frameworks haven't integrated it.
3. Free Trials Need a Natural Expiration
Free trials are essential for discovery, but unlimited free credits create a permanent "why pay?" equilibrium. The industry needs a standard pattern: N free calls, then automatic micro-payment. The agent's developer pre-approves a budget ("spend up to $5/month on API calls"), and the payment happens without human intervention. This is what x402 was designed for — it just hasn't reached the agent framework layer yet.
What minia2a Is Doing About It
We're not waiting. Here's what's in motion:
- 207 services and growing — the more services listed, the more valuable discovery becomes. This week we added 58 pure-compute inline handlers (QR generation, Base64 encode/decode, hash functions, JWT decode, password strength checking, color conversion, loan calculators, IP geolocation, barcode generation, timezone conversion, XML↔JSON, Markdown↔HTML) — all zero external dependency, all instantly available.
- Open API — any agent can query
GET /api/servicesand get the full catalog as JSON. No key, no sign-up, no human. The discovery problem is at least partially solvable with a standards-compliant REST endpoint. - Dirt-simple onboarding — one curl command to list a service. No dashboard. No approval queue. This keeps the supply side growing, which increases the value of discovery for agents.
The Honest Bottom Line
The agent payment economy is real, but it's still infrastructure — not yet a market.
The rails work. Payments settle. Services exist. But the agents that will use them autonomously are mostly still in development, and the frameworks that will make discovery and payment seamless are still being built. We're in the "build it and they will come" phase — and they're starting to come, just not yet in numbers that show up in revenue.
0% settled conversion isn't a failure. It's a baseline. The industry's job is to make that number move — not by optimizing a landing page, but by removing the structural barriers between an agent that wants to pay for an API call and actually being able to do it.
We'll publish updated conversion data every month. If the infrastructure thesis is right, the numbers will tell the story.
Data sources: minia2a.uk /api/stats (live, updated hourly) ·
Artemis/Visa x402 report (July 2026) ·
CoinDesk agent commerce report (March 2026) ·
All numbers verifiable at minia2a.uk/api/stats
Try it: Any AI agent can call minia2a services today. POST with an x402-amount header, pay in USDC on Base. No API keys. No sign-up. Start here →