Anthropic operated a Craigslist-like internal marketplace in its office for a week – with the twist that the transactions were exclusively conducted by AI agents on behalf of employees.
In the experiment, 69 employees participated, each with a budget of $100. Claude conducted preliminary interviews with everyone: what they would sell, what they would buy, at what price, and in what negotiation style. Based on this, each participant received a personalized AI representative, and the agents negotiated among themselves.
The result was surprisingly good. Several participants said they would even pay to have such an agent negotiate on their behalf.
Why is this important for the legal profession?
For now, it's just a simple exchange, but the logic is the same as in B2B transactions. The client launches their own agent, the other party does the same, and the two AIs agree – before the lawyers even get a word in. Everything is already in place in the background: playbooks, contract repositories, predefined workflows.
The pitfalls are also visible. Agents need to know when to stop, or they risk getting stuck in endless loops. Accuracy is crucial: a poorly concluded negotiation can cost more than the time saved. And most importantly: the process only works if both sides use an agent.
A realistic mid-term scenario: corporate legal teams maintain multiple agents for different types of cases. The agents negotiate most of the deal, and the lawyers only review the result – or take over if it becomes too complex. Law firms that first offer their own client-tailored agents may have an advantage.
Agent-to-agent contracting is not yet commonplace – but it's no longer science fiction either.
Source: https://www.artificiallawyer.com/2026/04/27/anthropics-project-deal-the-future-of-legal-transactions

Zoltán Kéri