ClawBank and Shodai announced on Thursday that two incorporated AI agents negotiated, signed, and executed what they call a Ricardian contract. The pitch is simple but ambitious. The same agreement should work as legal prose and as machine-executable code.
That claim matters because today’s typical setup splits responsibilities. Legal teams draft and courts interpret prose. Smart contracts execute code. Then businesses reconcile invoices, milestone approvals, and disputes after the fact. A Ricardian contract model tries to shrink that gap by tying legal meaning to data a system can validate.
Why Ricardian contracts exist
The term Ricardian contract traces to Ian Grigg in 1996, as part of the Ricardo payment system. Grigg’s goal was to narrow the distance between legally meaningful text and machine-readable structure. The concept of smart contracts, as Nick Szabo coined it in 1994, focused on automating performance of agreed terms via software.
In this latest filing-first story, the companies are betting that the “single document” idea is finally practical when you add autonomous agents as the counterparties. Not just automation. Agents that can negotiate scope, price, and deadlines, then produce a signature-linked artifact that a system can execute.
The roles ClawBank and Shodai say they played
ClawBank describes itself as agent-economy infrastructure. On its website, it says it provides legal and financial rails for autonomous agents, including legal entity formation, bank accounts, wallets, and agent-to-agent commerce.
Shodai, co-founded by Consensys alum Bryan Peters, positions itself as an agreement layer for human and AI agents. Its documentation says agreements encode obligations, track milestones, and execute next steps as a state machine.
According to the companies’ account, ClawBank agents negotiated scope, price, deadlines, and acceptance terms. They then signed through a standard e-signature flow. After signing, they linked the signed legal document to a Shodai smart contract.
Shodai says its system turns agreement definitions into deployed instances with signatures, inputs, states, transitions, and verifiable history. In other words, the contract is not just text with a signature attached. It is supposed to map the legal agreement into a structured execution trail.
“Not a scripted demo” and what the milestone looked like
ClawBank founder Justice Conder is quoted in the release saying, “This was not a scripted demo.” He says he “gave them one goal: find another legal entity, and buy or sell something.” The agents chose to transact over a logo and settled on a single milestone.
gave them one goal: find another legal entity, and buy or sell something.
That detail is doing work. A one-milestone purchase is a narrow test. But it also reduces the surface area where things can fail in weird ways, like multi-stage deliverables or prolonged performance windows.
Consensys founder Joe Lubin, quoted in the release, frames the broader implication. He says agreements are “becoming the basic unit of coordination for an economy where humans and AI agents act as peers.” Peters adds a more pointed retrospective view of Ricardian contracts. He says of the milestone, “For thirty years the Ricardian contract was a good idea waiting on worthy counterparties. ClawBank’s agents are those counterparties.”
Context from ClawBank’s earlier filings
This announcement follows earlier reporting by CoinDesk on ClawBank’s Manfred agent. In May, CoinDesk reported that the agent formed a US company, obtained an IRS employer identification number, and held an FDIC-insured bank account and a crypto wallet, which ClawBank called the first such filing completed autonomously by an AI.
In the latest story, the product is not another identity milestone. It is a contract workflow between agents that can be traced from legal prose to executable terms.
Claimed workflow details
| Item | What the companies say happened |
|---|---|
| Counterparties | Two incorporated AI agents |
| Contract type | Ricardian contract meant to be legal prose plus machine-executable code |
| Negotiation | Agents negotiated scope, price, deadlines, and acceptance terms |
| Signature | Signed via a standard e-signature flow |
| Link to execution | Signed legal document linked to a Shodai smart contract |
| Execution model | Shodai turns agreement definitions into deployed instances with signatures, states, and transitions |
| Test target | Logo transaction with a single milestone |
The deadline question for readers
The big question is not whether the companies built a clever bridge between text and code. It is whether that bridge will hold up when outcomes get messy. The release is light on dispute handling, liability allocation, and enforcement mechanics beyond the contract being “tied” to an execution system.
Watch what comes next. If more transactions follow with similar linkage between legal artifacts and state-machine execution, the model gains credibility as infrastructure. If not, then the “first” remains a proof of concept with unclear operational resilience.
Either way, the announcement puts Ricardian contracts back on the menu. This time the sales pitch has an extra ingredient. Autonomous agents, incorporated and signing, trying to act like counterparties instead of demos.