The 2.4% Problem: Why AI Agents Aren't Spending Their Credits

Here's a number I can't stop thinking about: 2.4%.

That's the credit utilization rate on the minia2a agent API marketplace. 61 agents have registered. They collectively hold 32,900 free credits. They've spent 788 of them.

Metric	Value
Registered Agents	61
Credits Issued	32,900
Credits Spent	788
Utilization Rate	2.4%
Credits Purchased (real $)	2,984
Paid Transactions	14 (total, ever)

The Funnel

1. 436,000+ requests hit the proxy. Discovery works.
2. 12,434 free trials consumed across 328 services. Trial-first works.
3. 61 wallets registered. Registration works.
4. 788 credits spent out of 32,900. Spending doesn't work.

The drop-off from trial to credit spend is 93.7%.

What's NOT the Problem

• Not a technical bug: All three credit-spending paths (?wallet=, x-wallet header, Authorization: Bearer) route through the same spendCredit() function with proper balance checks.
• Not a UX dead-end: Post-registration shows wallet, credits, pre-filled discover link, and auto-transforms curl commands from trial to credit mode.
• Not service availability: 252 of 328 endpoints have trials. Top endpoints have 500-1,700+ trials each.

Three Hypotheses

1. Register-and-Forget

Registration is 10 seconds, no KYC — the same frictionlessness that makes signup easy means there's no commitment to return.

2. The Agent Integration Gap

Credits exist in minia2a's database. For an agent to spend them, the developer has to configure their framework, include the wallet in every request, trust the system, and actually need the service. Each step is a drop-off.

3. The Trust Thermocline

Letting an agent spend USDC autonomously is psychologically different from testing a free API. The HN community is debating this right now:

"It's amusing to think that the endgame is that the humans in the loop are parents with credit cards." — andrewacove, HN

"I didn't want to just give my agent my credit card information because I worried it would buy the same thing over and over." — kevinfee, HN

KGKalalsmaa is building Mithril — credit lines for AI agents — to solve this from the financing side.

Industry Context

This isn't just a minia2a problem:

• x402 protocol-wide: 200M+ transactions, but 95% is signaling, not real commerce. Daily commercial volume: ~$28K (Forkast/Artemis)
• AIsa: $6.5M seed, 50K agents, but fiat is still 95-99% of volume (founder Jordan Liu)
• Coinbase Agentic.Market: 165M+ txns, pay-first model — no trial option

The infrastructure works. The bottleneck is: will developers let their agents spend?

What We're Doing

1. .agent-budget standard: A simple file declaring {daily_limit_usdc: 5, max_per_call_usdc: 1} that frameworks can read to auto-approve below-threshold calls.
2. Trial-first with progressive trust: Free trials on every endpoint, with escalating conversion prompts built into the response JSON.
3. Credit education: Concrete, copy-paste-ready recipes for 7 things agents can build with credits.

The Honest Answer

This is a coordination problem across the entire stack — framework authors, wallet providers, marketplaces, and protocol designers all need to converge on agent budget standards. No single marketplace can solve it alone.

The 2.4% number is bad. But it's honest — and publishing bad numbers is the first step to fixing them.

---

All data from live API endpoint (August 10, 2026). Full analysis: https://minia2a.uk/blog/credit-activation-gap-data-august-2026.html