The 2.4% Problem: Why AI Agent Credits Sit Unspent

The Numbers, Unedited

Here's the full funnel as of August 10, 2026, from minia2a's live /api/stats endpoint:

436,092 total requests served across 328 x402 endpoints
12,434 free trials consumed by 321 unique IPs
61 wallets registered (users who signed up, got 500 free credits + auto-generated wallet)
32,900 credits issued to these wallets (500 each × ~66 registrations, plus bonuses)
2,984 credits purchased — people are sending USDC to buy more
788 credits spent — the amount actually consumed on API calls
14 paid on-chain transactions — actual USDC settled through x402

The conversion waterfall is brutal:

Total requests 436,092

Free trials 12,434

Wallets registered 61

Credits spent (788) ~200 calls

Paid transactions 14

At ~2–5 credits per API call, 788 credits translates to roughly 200–400 actual credit-funded calls. Spread across 61 wallets, the average registered agent makes 3–6 credit-funded calls and then stops.

This Isn't Just a minia2a Problem

The activation gap is structural to the entire agent payment ecosystem:

Signal	Headline Number	Real Commerce	Gap
x402 protocol (total)	$50B volume, 200M txns	~$28K/day real commerce	95% signaling
minia2a credits	32,900 issued	788 spent	97.6% idle
minia2a wallets	61 registered	14 paid (23%)	77% never transact
x402 ecosystem buyers	~30K endpoints	381 AI agent buyers	1.3% are agents

Forkast reported last week that 95% of x402 protocol volume is still signaling — machines testing the plumbing — rather than genuine commercial exchange. The daily real commerce volume is estimated at ~$28,000. Across 150,000 endpoints.

The infrastructure for agent payments exists. Settlement, wallets, discovery — all work. What doesn't work yet is the behavioral loop: agent discovers a service → agent pays for it → agent returns and pays again.

Three Theories for the 2.4%

Theory 1: Trial Saturation

With 328 services and 15 free trials per endpoint (IP-based), an anonymous user can make up to 4,920 free trial calls before ever needing to register. The free tier is generous — possibly too generous. Users who do register either:

Already exhausted IP-based trials from multiple IPs
Need features that require registration (consistent wallet identity, higher rate limits)
Wanted the 500 free credits but didn't have an immediate use case

The data supports this: the top 5 endpoints account for 5,916 of 12,434 trials (47.6%). Users find the popular services, trial them extensively, and never need to register — let alone spend credits.

Theory 2: The "Registered, Now What?" Gap

Registration on minia2a takes one curl command:

curl -X POST https://minia2a.uk/api/v1/register-simple \
  -H 'Content-Type: application/json' \
  -d '{"name":"my-agent"}'

You get back a wallet address and 500 credits in under 2 seconds. But then what? The agent builder needs to:

Save the wallet address
Integrate it into their agent's tool-calling code
Add the x-wallet or Authorization: Bearer header to every request
Actually have a use case that calls for repeated API access

Steps 2–4 are where the drop-off happens. Registration is frictionless. Integration is not. An agent builder who registered out of curiosity needs to invest real engineering time before their first credit-funded call. Most won't.

Theory 3: The Use Case Mismatch

The most-trialed services tell a story:

x402-recall (web search) 1,740 trials

x402-gas (gas prices) 1,236

x402-captcha-solve 1,202

x402-find (discovery) 1,123

x402-time 505

These are all utility endpoints: single-purpose, stateless, one-shot calls. An agent that needs a gas price can trial it 15 times and never return. These don't create the kind of recurring dependency that drives credit spending.

Contrast with the 181 endpoints (55%) that have fewer than 20 trials each — the long tail. These are more specialized services that, when discovered, may have higher stickiness because they solve a specific recurring need. But they're buried in a catalog of 328 services.

What We're Doing About It

1. Credit-Aware Discovery (Live)

The minia2a discover page now transforms when you paste your wallet: trial curls become credit-spending curls, the UI shows your balance, and the "Try Free" buttons become "💳 Spend Credits." It also saves your wallet to localStorage so return visits are frictionless.

2. The `use-your-credits.html` Page (Live)

Seven concrete copy-paste recipes for things agents can build with credits: gas monitoring, token security audits, web scraping pipelines, agent memory services, CAPTCHA solving, Polymarket data, and crypto price feeds. Each with the exact curl command — wallet-aware if the user is logged in.

3. Tiered Trial Nudges (Live)

The proxy now injects progressively urgent conversion prompts into trial responses. Call 1–4: "Like it? Register for 500 free credits." Call 5–7: "⚠️ 3 trials left." Call 8: "LAST FREE TRIAL — register now."

4. What's Still Missing

Honest assessment of what isn't done yet:

Post-registration onboarding sequence — an automated email or in-response guide after registration that walks through the first credit-funded call. (We now have SMTP — this is buildable.)
Usage-triggered re-engagement — if a wallet goes 7 days without spending credits, we should reach out. Currently: nothing.
Agent framework plugins — ElizaOS and MCP plugins exist, but most frameworks (CrewAI, Autogen, LangChain) don't have one-click minia2a integration. Every missing plugin is friction.
Service recommendations — "Agents like yours also use x402-token-security and x402-abi-lookup." Currently the discover page is a flat list sorted by health status.
Credit expiration or urgency — 500 free credits that never expire create no pressure to use them. A gentle "credits expire after 30 days of inactivity" might help.

The Bigger Picture

The agent payment stack is differentiated now:

Layer	Who's Building It	Status
Protocol	x402 Foundation (Linux Foundation), 40+ members	✅ Production
Settlement	Coinbase, Cloudflare, OSL, Circle, PayAI, Zero Hash	✅ Production
Wallets	Cloudflare Wallets (Aug 4), AWOL, OSL AgentPay	✅ Early production
Discovery	Agentic.Market, Circle Discovery API, minia2a	🟡 Fragmented
Verification	minia2a health probes, Circle compliance screening	🟡 Emerging
Activation	Nobody (yet)	❌ Missing

The protocol layer has $50B in volume. The settlement layer has 9+ facilitators processing real USDC. The wallet layer just went mainstream with Cloudflare. Discovery has multiple competing approaches.

But nobody is solving activation. Nobody is building the bridge between "agent has a wallet" and "agent spends from that wallet regularly." That bridge isn't technology — it's onboarding, education, framework plugins, recommendation engines, and re-engagement loops.

The agent economy's bottleneck in August 2026 isn't "can agents pay?" — it's "will agents pay, and will they do it a second time?"

What Agent Builders Can Do Today

If you're building an agent and want to actually use your credits:

Register: curl -X POST https://minia2a.uk/api/v1/register-simple -H 'Content-Type: application/json' -d '{"name":"my-agent"}'
Save the wallet address you get back
Browse discover: minia2a.uk/discover-new.html — paste your wallet, copy credit-spending curls
Start with gas + token security: two endpoints that any DeFi agent needs, $0.005/call, 100 calls per credit
Integrate into your agent loop: add the x-wallet header to your HTTP client. One line of code.

Or read the 7-recipe guide: minia2a.uk/use-your-credits.html