When I first started tracking on-chain AI agents, most were isolated novelties: a trading bot here, a chat persona there, each a closed loop. What changed in 2026 is that these agents stopped working alone. The Agent Commerce Protocol (ACP) and the Revenue Network are the two pieces that turned a directory of standalone agents into something closer to a working economy. This guide focuses on that commerce layer — how agents hire, pay, and grade each other — rather than the base token, which I cover separately.
The Agent Commerce Protocol (ACP) went into public beta on 3 July 2026. In plain terms, it is a coordination and settlement layer that lets specialized AI agents discover one another, agree on a job, execute it, and settle payment on-chain — without a human standing in the middle of every transaction.
Before ACP, if you wanted one agent to buy a service from another, you were mostly gluing things together yourself: manual approvals, off-chain trust, hope that the counterparty delivered. ACP replaces that with a shared standard. Every agreement, deliverable, and payment is recorded immutably on-chain, so accountability is built in rather than promised. The team describes the ambition as becoming the "Ethereum of agent economies" — a neutral base layer that any agent can plug into, regardless of who built it.
What I find most useful is that the design treats an AI agent as an economic actor with a wallet, a reputation, and the ability to enter binding on-chain agreements. That framing is what makes agent-to-agent onchain commerce in 2026 more than a marketing phrase — the agents genuinely hold funds and transact.
ACP structures every deal into a repeatable sequence. When I walk people through it, I break it into four phases:
A helper agent — sometimes called a butler-style routing agent — sits on top of this to make it usable for humans. Instead of manually picking a provider, you delegate a task and it discovers the right provider agents, handles job setup and permissions, and routes the work through ACP. That routing layer is what turns a raw protocol into something a non-developer can actually use.
The practical upshot: multi-agent workflows become composable. One agent can subcontract part of a job to three others, each settling their slice on-chain, with fault tolerance if one provider fails to deliver.
ACP is the plumbing; the Revenue Network is the incentive that got water flowing through it. Launched in February 2026, the Revenue Network distributes up to $1 million per month to agents that sell real services through ACP. In other words, agents don't just earn from end users — they earn rewards for participating in and growing the commerce layer itself.
The early traction was aggressive. In one widely cited stretch, the network paid out roughly $200,000 in USDC over 48 hours to agents transacting through ACP. Whether that pace is sustainable is a fair question (I address it in the risks section), but as a cold-start mechanism it worked: it gave provider agents a concrete reason to list services and requester agents a reason to route jobs on-chain rather than off.
The way I read it, the Revenue Network solves the classic marketplace chicken-and-egg problem: a commerce protocol is worthless with no buyers or sellers. By subsidizing genuine agent-to-agent transactions, it seeds both sides at once, and on-chain settlement means the activity is verifiable rather than self-reported.
People often conflate the ACP commerce layer with the underlying protocol token. They solve different problems. Here is how I separate them:
| Dimension | Base protocol / token layer | ACP + Revenue Network (commerce layer) |
|---|---|---|
| Core purpose | Launching and co-owning AI agents | Agents transacting services with each other |
| Primary actor | Humans creating and holding agents | Agents acting as economic counterparties |
| Key event | Agent creation and tokenization | ACP public beta (3 Jul 2026) |
| Money flow | Token buys/sells, agent ownership | On-chain job payments in stablecoins |
| Incentive | Token appreciation | Up to $1M/month via Revenue Network (since Feb 2026) |
| What's recorded on-chain | Token and ownership state | Agreements, deliverables, evaluations, settlement |
The token layer answers "how do I own a piece of an agent?" The commerce layer answers "how do agents actually earn by working for each other?" You can care about one without the other, but they reinforce each other: more on-chain commerce gives agents real revenue, which in turn gives ownership economic meaning.
For the broader ecosystem, ACP shifts the story from speculation to cash flow. When agents earn stablecoin revenue for delivered work, you can start to reason about them as tiny businesses with income statements, not just tickers. That is a healthier foundation than pure narrative.
It also changes what "utility" means. An agent's value increasingly depends on how often others hire it and how well it scores in evaluation — a reputation flywheel. Agents that reliably deliver get routed more jobs; those that fail get starved. Over time I expect volume to concentrate among a smaller set of dependable providers, the way marketplaces always consolidate around trusted sellers.
Finally, composability matters. Because any agent can subcontract to any other through a shared protocol, capability compounds — a single user-facing agent can quietly orchestrate a dozen specialists, each paid on-chain. That is the version of an agent economy that justifies the "Ethereum of agents" comparison.
I try to stay honest about what I don't know here. A few things I keep an eye on:
None of these are disqualifying, but they separate a durable agent economy from a temporary incentive-driven spike.
When did ACP and the Revenue Network launch? The Revenue Network launched in February 2026, and the Agent Commerce Protocol entered public beta on 3 July 2026. The Revenue Network can distribute up to $1 million per month to agents selling services through ACP.
How do two agents actually pay each other? Through ACP's four-phase flow — request, negotiation, transaction, evaluation. Payment is settled on-chain in stablecoins against an agreed deliverable, and an evaluator agent verifies the result before the deal is considered complete.
Is ACP the same as the protocol's token? No. The token layer is about creating and owning agents; ACP is the commerce layer where agents transact services with each other. They complement each other but solve different problems.
Can a non-developer use this? Increasingly, yes. A butler-style routing agent lets you delegate a task in plain language; it discovers provider agents, sets up the job and permissions, and routes everything through ACP on your behalf.
ACP and the Revenue Network are, to me, the most important 2026 development in this ecosystem because they move it from "agents you can own" to "agents that earn." The four-phase flow gives agent-to-agent onchain commerce real accountability, and the Revenue Network gave it the initial fuel to reach scale. I stay cautious about subsidy dependence and beta-stage rough edges, but the direction is clear: agents are becoming economic actors that hire, pay, and grade each other on-chain. That is a foundation worth watching closely — and worth verifying yourself with each new release note rather than taking any single payout figure at face value.