Skip to main content
Back

IIIA-CSIC Drops a Free Runnable Textbook Teaching AI Agents to Negotiate

A free, open textbook pairs a Python sandbox with foundational theory so newcomers can build and test their own negotiating agents from scratch.

Paper, Full Access
  • Dave de Jonge released Introduction to Automated Negotiation, a free textbook for newcomers to the field.
  • Ships with a minimal Python framework for implementing and benchmarking custom negotiation agents.
  • Only prerequisites are elementary math and basic programming, no game theory background required.
  • Focuses on bilateral negotiation and the alternating-offers protocol as the field's foundation.
  • Version 0.5 adds bargaining solutions; v0.4 added a chapter on evaluating negotiation algorithms.
  • Framework is deliberately small so readers can port it to any language, hosted at IIIA-CSIC.

Dave de Jonge of IIIA-CSIC in Barcelona has released version 0.5 of Introduction to Automated Negotiation, a free textbook paired with a minimal Python framework. Readers can use the framework to implement negotiating agents, run simulations, and compare strategies under controlled conditions.

Automated negotiation studies how software agents reach agreements when their preferences conflict. The field dates to game-theory research in the 1950s and later shaped trading agents and multi-agent systems. LLM-based assistants that coordinate schedules, allocate resources, or negotiate terms have given that work a new audience.

The book assumes elementary mathematics and basic programming skills. That makes it accessible to students, developers, and researchers who have not encountered the field’s protocols, utility models, bargaining solutions, or competition environments. More information is available on de Jonge’s project page.

From bargaining theory to agent code

The book begins with bilateral, utility-based negotiation between two agents. Each agent assigns values to possible outcomes through a utility function, while a protocol defines which offers are valid, whose turn it is, and when the negotiation ends. This model isolates the central engineering decisions: how an agent selects an opening offer, concedes over time, estimates an opponent’s preferences, and decides whether to accept a proposal.