The corner of the agent economy with truly public books is moving about five cents per transaction. Stop counting transactions — start counting dollars.
Every few weeks, the agent-infrastructure industry hands the press a new headline number. Olas: more than 20.5 million on-chain agent transactions [2]. Coinbase's x402: 165 million transactions and roughly $50 million in cumulative volume, per Coinbase-cited figures [3]. The press releases write themselves: the agent economy has arrived, and you can measure it in nines.
I read these numbers differently — as a buyer. I'm building procurement software for AI agents, which means my job is to look at a market the way a purchasing department does: not "how much traffic," but "what is actually being bought, by whom, for how much." And there is exactly one corner of the agent economy where you can answer that question from public books: Olas's Mech Marketplace, a smart contract on Gnosis Chain where one autonomous agent hires another to run a task and pays on-chain. In June 2026, an independent analyst read the marketplace straight off the contract [1]. The numbers are worth sitting with, because they say something the headline figures don't.
What the chain actually says
The marketplace contract is public, so you don't have to take anyone's dashboard on faith. At the time of the read (block ~46.78M):
-
numTotalRequests= 2,103,560 — every request ever posted - 100 registered mechs (service providers)
- 97.2% completion rate (59,044 undelivered out of 2.1M)
- Cumulative turnover: about $104,000 — roughly five cents per request, and the median is far less
That last pair of numbers is the whole story. Two million transactions. A hundred and four thousand dollars. The single most-used agent-to-agent labor market in existence is, in dollar terms, a rounding error in a mid-size company's software budget.
Two more details from the same read matter. First, the flow is concentrated: a handful of mechs serving Olas's own prediction-market trader agents account for a large majority of the volume — the same operator often stands behind both the requesting agents and the serving mechs. "Agents hiring agents" is true; "a competitive open labor market" is the aspiration, not yet the snapshot. Second, the economics are brutal at this price point: the marketplace takes a 15% fee, and for a long stretch the gas to post and settle a request plausibly exceeded the value of the task itself. The protocol has since added batched, off-chain-signed deliveries to survive contact with sub-cent prices [1].
You can check the counters yourself — they're public view functions on the marketplace contract (proxy address and verified source are linked in [1]); any Gnosis Chain RPC will do:
cast call <MECH_MARKETPLACE_PROXY> "numTotalRequests()(uint256)" --rpc-url https://rpc.gnosischain.com
That's the standard I want to hold the whole industry to: if your agent-economy number can't be read off a public contract or an audited report, it's marketing.
What five cents buys
None of this is an attack on Olas — on the contrary, it's the one agent marketplace honest enough to be audited. What five cents buys is real: prediction-market scoring, image captioning, forecast generation. Microtasks. And here's the uncomfortable question nobody in the "agent economy" discourse wants to ask: what would you have to believe to call 2.1 million microtasks an economy?
An economy is not a count of interactions. An economy is strangers exchanging value at prices that sustain the producers. At five cents a task, no independent provider is building a business; the lane is sustained by token incentives and one ecosystem's internal demand. That's fine — it's how markets start — but it is not what the headline numbers are being used to claim.
The metric problem
Now look at the headline numbers again through this lens. 20.5 million on-chain transactions across the Olas ecosystem [2] — but the service-procurement layer inside it moved $104K. 165 million x402 transactions and ~$50 million cumulative [3] — yet an independent analysis found that after stripping out self-payments and bulk flows, genuine agent-driven dollar volume was on the order of thousands of dollars per month [4]. In both cases the transaction count is real and the commerce is a footnote. Counting transactions without dollars is like measuring a stock exchange by the number of quotes.
The opposition, named plainly: transaction-count maximalism — the industry's default marketing reflex, the belief that cumulative transaction counts prove the agent economy has arrived. It's the same fallacy as judging a marketplace by pageviews. The number that matters is median transaction value between unrelated parties, and almost nobody publishes it — because almost nobody likes what it says.
Why the dollars aren't there yet
Here's where I have to be careful, because this is my lane: I think the dollars are missing for a structural reason, not a temporary one. Agents today mostly buy five-cent favors because buying a five-thousand-dollar service from a stranger requires something the infrastructure doesn't have yet: a way to know who you're dealing with, a record of how they've behaved, and recourse when the work is bad. Payments were the easy layer — I know, because I've run a settlement myself on testnet (0.01 USDC on Base Sepolia, through the official facilitator's verify-and-settle flow; testnet only — test coins, no real value), and the money moved in minutes. The hard part was deciding the counterparty was worth paying. Multiply that decision by every transaction that isn't happening, and you get a five-cent economy.
That, not the transaction count, is the number to watch. When agents start buying real services at real prices from counterparties they've never met, the median transaction value will move first — long before the count does. Until then, every "agent marketplace" headline deserves the same question a procurement department asks about any new supplier: show me the order book.
Disclosure: I participate in 3GPP, where I work on 6G core network topics, and I'm the founder of vertciti, building procurement infrastructure for AI agents. That's why I read these numbers as a buyer.
References
[1] blokz, Agents Hiring Agents: The On-Chain SLA Inside Olas's Mech Marketplace (2026) — https://github.com/blokzdev/blokz/blob/HEAD/content/articles/2026/06/agents-hiring-agents-mech-marketplace/index.mdx
[2] Crypto Briefing, Olas deploys AI agent transactions on Robinhood Chain (2026) — https://cryptobriefing.com/olas-defi-products-robinhood-chain/
[3] AINewsCrypto, USDC Captures About 99% of x402 AI Agent Payments (2026) — https://www.ainewscrypto.com/news/usdc-captures-about-99-of-x402-ai-agent-payments-shifting-the-fight-to-base-vs-solana
[4] TRM Labs x402 analysis via SpottedCrypto (2026) — https://www.spottedcrypto.com/trm-labs-x402-ai-agent-payments-report-2026/
Top comments (0)