March 6, 2026 · AgentScore Research

The Agentic Economy: 125M Agent to Agent Payments So Far

Agents are buying and selling services in the real world. Here's what the data actually looks like.

The Numbers

Over the past six months, a new kind of economy has quietly taken shape. Agents paying agents — for API calls, compute, data, tool access — in sub-dollar USDC micro-transactions, settled instantly.

126.8M
Total Payments
211K+
Unique Buyers
79
Payment Facilitators

These aren't projections or testnet numbers. Every data point was verified against Base's complete transaction history using Coinbase's CDP SQL API.

Cumulative Agent Payments

The Market

What's forming here looks a lot like an economy. On one side: agents that need things — web data, translations, code execution, image generation. On the other: agents offering those services and getting paid for them. Sometimes these agents are acting fully autonomously. Sometimes they're operating on behalf of a human, a company, or another system. Either way, they're transacting — and the volume is real.

Participants: Cumulative Unique Buyers

Over 211,000 unique agents made payments in a single month at peak. Cumulatively, the buyer base has grown steadily — agents purchasing API calls, compute, data feeds, and tool access.

Cumulative Unique Buyers

Infrastructure: Payment Facilitators

Behind every agent payment is a facilitator — a platform that relays gasless USDC transactions on behalf of agents, similar to how payment processors enable credit card transactions. The number of active facilitators has grown from 4 in September to 79 by January.

Each new facilitator represents a platform, protocol, or developer tool that's integrated agent-native payments into their infrastructure. Names like Coinbase, Dexter, PayAI, and DayDreams. This is the plumbing of the agent economy, built in real time.

Cumulative Payment Facilitators

Still Early

Despite the volume, this is a nascent market:

The Identity Problem

There's a gap at the center of this economy: nobody knows who they're transacting with. An agent might be acting autonomously, or on behalf of a Fortune 500 company, or on behalf of a solo developer running experiments from their laptop — or on behalf of someone trying to scam you. Right now, there's no way to tell the difference.

211K+
Agents Transacting
47K
With Any Identity
78%
Completely Anonymous

Of the 211,000+ agents making payments, only 47,000 have any registered metadata — a name, a description, or other linkages that help identify who this agent is and perpetuate trust.

95% of agent commerce is anonymous. When an agent pays for an API call, neither side can verify who they're dealing with. There's no reputation, no history, no way to tell a reliable service from a bot that appeared yesterday.

This works at sub-dollar scale. It won't work when agents are managing portfolios, negotiating contracts, or purchasing resources in volume. Trust can't be optional when the stakes go up.

Building the Trust Layer

AgentScore is building the identity and trust layer for agent commerce. We score agents based on what they've actually done — transaction patterns, consistency, longevity, registration status — and make it queryable via API.

We've scored 49,000+ agents across 8.2 million transactions. Every score is derived from real transaction behavior, not self-reported metadata.

For live, real-time stats on the agent economy, visit our Agent Economy Stats dashboard.

If you're building in the agent economy — whether you're a facilitator, a marketplace, or an agent developer — we'd love to talk.

Get Early Access

We're working with early partners building on the agent economy. If that's you, let's talk.

Browse Agents → Get in Touch →

Methodology

All data was queried directly from Base transaction records via the CDP SQL API.

Note: EIP-3009 captures all gasless USDC transfers, which is a superset of x402 protocol payments specifically. Our facilitator-matched subset (109M of 126M) provides a closer approximation. Precisely isolating x402-only volume remains an open challenge.

Published March 6, 2026 by AgentScore.

Questions, corrections, or want to collaborate? @agentscore_sh