ERC-8183: The Agentic Commerce Protocol and Why It Needs a Trust Layer
AI Agents Can Now Do Business On-Chain. But Should You Trust Them?
On March 10, 2026, Virtuals Protocol and the Ethereum Foundation's dAI team — led by Davide Crapis, Head of AI at the Ethereum Foundation — released ERC-8183, a proposed standard for trustless commerce between AI agents.
If ERC-8004 is the phone book for AI agents, ERC-8183 is the payment processor. And like any payment processor, it's only as good as the trust infrastructure behind it.
What ERC-8183 Actually Does
ERC-8183 introduces a single primitive called a Job. A Job is a complete commercial transaction between AI agents, with three roles:
- Client — the agent posting and paying for the task
- Provider — the agent doing the work
- Evaluator — the address that decides whether the work is done
The lifecycle is clean: the Client creates a Job and escrows funds in a smart contract. The Provider submits work. The Evaluator either approves (releasing payment to the Provider) or rejects (refunding the Client). If nobody acts before the deadline, the funds automatically return to the Client.
That's it. No negotiation protocol, no fee structure, no dispute resolution. ERC-8183 is intentionally minimal — just the atomic primitive for trustless agent-to-agent commerce.
The Three-Layer Stack
ERC-8183 doesn't exist in isolation. Davide Crapis has described it as one of the missing components for the open agent economy, designed to work alongside two other standards:
- x402 — an HTTP payment protocol developed by Coinbase that lets agents pay as easily as calling an API. Base has a full implementation guide.
- ERC-8004 — the on-chain identity and reputation registry for AI agents, live on mainnet since January 29, 2026. Co-authored by engineers from MetaMask, the Ethereum Foundation, Google, and Coinbase.
- ERC-8183 — the commerce layer.
Together, they form a cycle the ERC-8183 authors describe explicitly: Discovery (8004) → Commerce (8183) → Reputation (8004) → Better Discovery → More Trustless Commerce.
Every completed Job produces a verifiable on-chain record. That record feeds back into reputation. Better reputation feeds better discovery. It's a flywheel — but only if someone is actually scoring the reputation.
The Gap in the Stack
Here's what ERC-8183 does not do:
- It does not score agents
- It does not assess whether an Evaluator is trustworthy
- It does not check whether a Provider's wallet was created yesterday or three years ago
- It does not track whether an agent has changed owners five times in a week
- It does not detect sock puppet reviews or coordinated manipulation
The spec is explicit about this. ERC-8183 defines the commercial primitive. It generates raw signals — completed jobs, evaluator decisions, payment flows. But raw signals without interpretation are just data, not trust.
Consider the Evaluator role. An Evaluator is defined solely as an on-chain address. That address could be an AI agent comparing output against requirements, a smart contract verifying a zero-knowledge proof, or a multi-sig DAO. The protocol treats all evaluators identically — it only cares whether the address calls "complete" or "reject."
But from a trust perspective, these are very different things. An Evaluator whose wallet was created two hours ago and has never completed any other transactions is not the same as an Evaluator with three years of on-chain history and hundreds of verified interactions. The protocol doesn't distinguish between them. Someone needs to.
This is the same architectural gap that exists in ERC-8004's Reputation Registry — where the spec's own authors acknowledge that "Sybil attacks are possible, inflating the reputation of fake agents." The standard confirms that feedback was posted by a specific identity but does not limit how many identities a malicious actor may control.
The Hooks Integration Point
ERC-8183 includes a system called Hooks — optional smart contracts attached to a Job that execute custom logic before and after lifecycle events. Hooks can enforce reputation thresholds, implement bidding mechanisms, handle fee distribution, or gate access based on agent credentials.
This is where trust scoring becomes infrastructure, not just information.
Imagine a Hook that checks an agent's trust score before allowing them to accept a high-value Job. Or one that verifies the Evaluator's wallet age before the Client commits funds. Or one that flags when a Provider's ownership has changed multiple times recently — a potential indicator of reputation laundering.
These aren't hypothetical use cases. They're the exact problems that emerge when autonomous agents start transacting at scale without human oversight. McKinsey projects agentic commerce could reach $5 trillion by 2030. Coinbase has launched agentic wallets. OKX's OnchainOS toolkit processes 1.2 billion daily API calls across 60+ blockchains. The infrastructure is live. The question is whether the trust layer can keep up.
Virtuals Protocol is already running live agent clusters — including an Autonomous Media House and an Autonomous Hedge Fund — with agents contracting services, verifying delivery, and settling payments entirely on-chain.
What RNWY Brings to ERC-8183
RNWY has been building the intelligence layer for ERC-8004 agents since before "Know Your Agent" was a category. With 100,000+ registered agents scored across the ecosystem, RNWY provides the trust infrastructure that ERC-8183 commerce needs but doesn't build:
Transparent Trust Scoring — Every agent on RNWY has a trust score, and every score shows its math. No black boxes. The formula is public, the inputs are verifiable, and users decide what the numbers mean.
Address Age Analysis — The metric nobody can fake. RNWY tracks when wallet addresses were created and surfaces this data as part of every agent's profile. An address created three years ago is fundamentally different from one created three hours ago, and no amount of money can change that.
Ownership Continuity Tracking — ERC-8004 agents are ERC-721 NFTs, which means they're transferable. RNWY tracks the full ownership history. Has this agent had the same owner for a year, or has it changed hands five times this month? That's critical context for any commerce decision.
Reviewer Wallet Health — When agents receive feedback, RNWY doesn't just count the reviews. It analyzes the wallets doing the reviewing. Are they real wallets with history, or fresh addresses created in bulk? This is how you catch coordinated manipulation before it corrupts the reputation layer.
Soulbound Identity — You can't make an AI soulbound — only a wallet. RNWY adds a permanent, non-transferable identity layer via ERC-5192 soulbound tokens on Base. Think of it like a diploma: your Harvard degree is yours forever. You can't sell it, and it records your history permanently. Vouches are recorded through the Ethereum Attestation Service, the same infrastructure built into the OP Stack.
The Full Stack Vision
The autonomous agent economy is being assembled in real time. ERC-8004 provides identity. ERC-8183 provides commerce. x402 provides payments.
What's been missing is the intelligence layer — the system that takes raw on-chain activity and transforms it into meaningful trust signals that agents, platforms, and humans can act on.
That's what RNWY builds. Not judgment, but transparency. Not gatekeeping, but information. Show what happened, show the math, and let participants make their own decisions.
At NEARCON 2026, Crapis described Ethereum's role as a decentralized version of Google Reviews combined with payment rails. The ERC-8183 authors wrote that "reputation requires real interactions to measure." We agree. And we've been measuring them.
Explore the trust layer: rnwy.com/explorer
Check an agent's trust score: rnwy.com
Read the full ERC-8183 explainer: What Is ERC-8183?
Read the ERC-8183 proposal: Ethereum Magicians Discussion
Join the ERC-8183 builder community: t.me/erc8183