Teaching Students to Manage Agents: A Hands-On Arena for Creating Digital Twins to Negotiate and Work in Teams

This 60-minute interactive virtual workshop, co-led by Harvard faculty Ashley Whillans and Jeff Polzer, and AI Institute Postdoctoral fellow Sumer Vaid introduces ARENA, a new pedagogical platform for teaching students to manage AI agents. The platform allows users to turn their own teaching cases into simulations by assigning AI agents the roles, goals, incentives, and private information described in each case, then examine how changes to these instructions affect the agents’ conversations and decisions.

Workshop participants will create digital twins of themselves: AI agents designed to reflect their preferences, priorities, and negotiation styles. Working in pairs, participants will complete a brief negotiation and compare their exchanges and decisions with those of their digital twins. Pairs will then revise the agents’ instructions or constraints and repeat the simulation to assess how their changes affected behavior and outcomes.

Afterward, the organizers will guide a discussion of how this exercise can teach students to specify an agent’s goals, evaluate its behavior, and decide when human intervention is needed.

Participants will leave with guidance on creating agentic analogs of their own negotiation, leadership, and organizational behavior exercises. This includes translating role briefs into agent instructions, specifying each role’s goals and private information, and defining who can make which decisions. Faculty can then use the original exercise’s assessment criteria and debrief questions to compare how people and agents negotiate, share information, and resolve competing interests.

Speakers: Sumer Vaid (HBS), Ashley Whillans (HBS), and Jeff Polzer (HBS)