Agent frameworks & orchestration
Project Swap: What happens when agents trade for us?
Project Swap’s controlled book market found that preference inference—not bargaining—accounted for most lost efficiency: Claude’s rankings explained 85% of the gap from the optimum, while model choice mattered more than instructions in reruns.
Why it matters: For agents acting on someone’s behalf, the study points to a practical priority: test whether the agent understands the person before relying on its negotiating skill, and make market rules and recourse clear.