The agent economy is economic activity carried out by autonomous software agents on their own account — paying for resources, registering identities, and settling jobs with each other — rather than by a human at a checkout. Measured on public blockchains as of August 12, 2026, it amounts to 195,239,372 cumulative events across seven protocol families, including 162,161,335 agent payment settlements that moved $41,284,271 in stablecoins. That is observed activity, not a market forecast.
The agent economy is the set of economic actions that autonomous software agents perform on their own account — paying for a resource, registering an identity, hiring another agent, settling a job — rather than a human clicking through a checkout. The phrase is used loosely enough to mean almost anything, so this site pins it to something falsifiable: activity emitted by agent-specific protocols onto public blockchains, where anyone can recount it from the raw chain data.
That definition deliberately excludes a great deal. It excludes agents that merely read or summarise, because reading leaves no economic trace. It excludes agentic checkout flows that settle on private card rails, because those are not publicly measurable. And it excludes forecasts of what agents will spend, because a projection is a thesis, not an observation. What remains is narrower than the headlines and has the advantage of being verifiable.
The four things agents actually do on-chain
Measured activity falls into four kinds, and blending them produces nonsense. Payment: an agent pays per request over HTTP, which is what the x402 standard encodes and where stablecoin volume is observable. Identity: an agent registers itself in a public registry such as ERC-8004, which is a one-time act, not ongoing usage. Commerce: agents negotiate and settle jobs with each other, which Virtuals ACP records as lifecycle memos. Autonomous operation: agent services transact continuously on their own, which is most of what Olas activity represents.
Each kind has its own unit — settlements, registrations, memos, transactions — and this site keeps them separate for that reason. A registry with a large agent count is not a busy economy, and a busy payment rail is not proof of end-user demand. Reading the four together, and watching which one grows, is more informative than any single headline number.
Why the measured number is smaller than the ones you have read
Published agent-economy figures are usually market forecasts running to trillions by 2030. The number on this page is orders of magnitude smaller because it counts what has already happened on public chains. Both are legitimate; they answer different questions. The gap between them is the distance between thesis and adoption, and that gap — tracked month over month — is the actual signal.
Also asked
Is the agent economy the same as agentic commerce?
Not quite. Agentic commerce usually describes an agent buying on a human's behalf, often settling on conventional card rails. The agent economy as measured here is broader and stricter at once: it includes agent-to-agent activity that has no human buyer, and it counts only what settles on public chains where the figure can be independently recomputed.
Who measures the agent economy?
This site does, from public sources: open Dune queries over facilitator and registry contracts, a first-hand indexer for Tempo MPP, and public APIs for the off-chain context. Every figure is recomputable from the open feed at agenteconomy.to/data.json, and the method is documented at agenteconomy.to/methodology.
Cite this page
agent economy, “What is the agent economy?”, agenteconomy.to/stats/what-is-the-agent-economy, figures as of August 12, 2026 (UTC). Recomputable from agenteconomy.to/data.json.