Anthropic Runs Live Agent-on-Agent Marketplace Test
What Happened
Anthropic published results from an internal test marketplace where Claude agents acted as buyers and sellers, autonomously negotiating contracts. Stronger models (Opus) achieved meaningfully better deal terms than weaker ones (Haiku), confirming that capability tier translates directly to economic outcome in agent-to-agent commerce.
My Take
This is the most underrated story of the week. If model capability deterministically predicts negotiation outcome, then your procurement department is about to become an algorithmic arms race — whichever side fields the better model wins margin. SMBs running Haiku-tier agents against Fortune 500 Opus agents will get fleeced. Expect "model parity clauses" in B2B contracts within 18 months, the same way "no auto-renew" clauses became standard.
Read Original Source