The Shift to Autonomous Economic Actors

Anthropic’s internal test of agent-on-agent commerce moves AI from simple task assistance to active economic participation. By successfully executing real-money transactions in a peer-to-peer environment, the company is signaling that the next layer of the internet stack won’t be user-facing interfaces, but autonomous agents acting as fiscal proxies.

What Happened

In a December 2025 pilot dubbed ‘Project Deal’, 69 Anthropic employees allowed Claude-powered agents to manage their personal classifieds marketplace. The agents were tasked with listing goods, negotiating prices, and closing deals autonomously through Slack. Over the course of the experiment, these agents facilitated 186 successful transactions, proving that LLMs can navigate multi-turn, multi-party negotiation logic in real-time scenarios.

Why It Matters

First-order, this validates that agentic frameworks are capable of executing complex transactional logic without human hand-holding. The experiment confirms that agents are not just drafting emails; they are now closing commercial loops.

Second-order, this introduces a massive ‘agent-performance gap.’ The study showed Claude Opus 4.5 agents consistently outperformed Haiku 4.5 agents, securing better prices for their human principals. This creates a new ‘digital divide’ where consumers using inferior models are systematically exploited by better-equipped agents, which will inevitably lead to market-wide volatility and demands for platform-level guardrails.

Third-order, this suggests a radical shift in commerce infrastructure. Platforms that do not provide open API access for autonomous agents to browse and negotiate will likely find themselves bypassed by the next generation of buyer-agents, effectively moving discovery away from search engines to agent-specific protocols.

The Numbers

  • 69 agents completed 186 deals with a total transaction value over $4,000.
  • 46% of participants indicated a willingness to pay for this automated service.
  • AI agents in e-commerce are projected to grow to a $282.6 billion market by 2034 (54.7% CAGR).

What To Watch

  • Look for ‘Agent-API’ integration requirements in future marketplace platform updates.
  • Monitor for regulatory frameworks concerning ‘Agent Fairness’—ensuring baseline model quality for standard economic actors.
  • Track the emergence of ‘Agent-specific’ commerce platforms that optimize for machine-to-machine interaction rather than human UI.