Circle Just Launched an Agent Discovery API. Here's What It's Missing.

On August 6, Circle launched a Discovery API — a public catalog of 900+ USDC-compatible service endpoints that AI agents can query programmatically. No API keys. No authentication. Just a machine-readable directory of paid services, filterable by category, network, and price.

This matters because Circle is the company behind USDC, which settles 98.6% of all agent-to-agent payments. They already control the settlement rail. Now they're building the discovery layer on top of it. If they get both right, they own the full stack — from "agent finds a service" to "USDC lands in the provider's wallet."

But here's what Circle didn't build: a trial system.

The Catalog Works for Humans. Agents Need Something Different.

A searchable catalog of 900 services with names, categories, and prices is useful — for a human developer picking tools. But the agent economy isn't humans browsing a catalog. It's agents making sub-second decisions about which endpoint to call.

Here's the real discovery flow for an agent:

1. Agent encounters a task — "I need the current gas price on Base"
2. Agent queries a discovery API — "show me endpoints that return gas prices"
3. Discovery API returns 4 options — GasNow (0.01 USDC), GasTracker (0.02 USDC), FeeOracle (0.015 USDC), BaseScan Gas (0.01 USDC)
4. Now what?

At step 4, the agent has a list of names, descriptions, and prices. But it has zero information about which one actually works. Which one returns valid JSON? Which one has 200ms latency vs 2-second timeouts? Which one's response format matches what the agent needs to parse?

A human would open each in a browser tab, read the docs, maybe run a curl command. An agent can't do that. An agent needs to call the endpoint to evaluate it. And if every evaluation call costs money, discovery becomes expensive before it becomes useful.

The Trial Gap Is a Structural Problem

This isn't a missing feature. It's a missing layer of the stack.

The agent payment stack has four layers, and the industry has only built three:

Circle's Discovery API is layer 3 — an excellent catalog, well-designed, backed by the dominant settlement provider. But layer 4 is where the agent actually decides. And without a trial layer, the catalog is a phone book where every call costs money.

Why Nobody Builds Trial Infrastructure

It's not an oversight. Building a trial system for agent-to-agent payments is genuinely hard:

• Abuse prevention: If trials are free, how do you prevent an agent from consuming 10,000 free calls across 900 endpoints? You need per-endpoint rate limiting, not just per-user.
• Provider incentives: Why would an API provider offer free trials to unknown agents? The trial system has to make the economics work for the provider, not just the consumer.
• Credit accounting: Issuing trial credits costs money if those credits can be converted to real usage. The accounting layer has to track trial vs. paid usage separately, with clean transitions.
• Agent identity: How do you associate trials with a specific agent wallet without requiring KYC? The trial system has to work at machine speed with machine identity.

Circle could build this. They have the engineering resources, the USDC infrastructure, and the regulatory framework. But they chose not to — at least in v1. Their Discovery API is a catalog. It answers "what's available?" It doesn't answer "what works for me?"

The Trial-First Architecture

The alternative is an architecture where every endpoint in the catalog can be called with a trial flag before any payment is made.

# Agent discovers endpoints
GET /api/discover?category=gas

# Agent trials the top result before committing budget
POST /api/gas
Authorization: Bearer <wallet-key>
X-Trial: true
→ {"gas_price": "0.12", "source": "base", "trial": true}

# If the response format works, agent pays for subsequent calls
POST /api/gas
Authorization: Bearer <wallet-key>
X-Payment: x402
→ {"gas_price": "0.12", "source": "base"} + HTTP 402 receipt

This isn't theoretical. The ?trial=1 pattern works in production today. 11,640 trials have been served across 324 endpoints, with per-endpoint caps that prevent abuse. The agent gets to validate the endpoint before spending. The provider gets discovery without giving away unlimited free API calls. The marketplace gets conversion data: which trials lead to paid usage, and why.

Circle's catalog is an important step forward for the agent economy. USDC settlement plus programmatic discovery is the right foundation. But until the trial layer exists, agent discovery will be a human-mediated process — developers browsing a catalog and hardcoding endpoint choices — rather than something agents can do autonomously.

What Happens Next

The next 6 months will determine whether discovery consolidates around a single provider (Circle's USDC monopoly + their catalog) or fragments across multiple discovery layers differentiated by features above the catalog — trial infrastructure, reliability scoring, response format validation, and auto-delisting of dead endpoints.

Circle owns the settlement layer. But the discovery layer is still a race. And the winner won't be the platform with the most listings. It'll be the platform where an agent can answer the only question that matters: "which of these endpoints should I actually call?"

Bottom line: Circle built a great catalog. But catalogs are for humans. Agents need test drives. The discovery platform that solves trial infrastructure wins the next phase of the agent economy.