AI agents are no longer limited to answering questions. They can use tools, consult documents, coordinate tasks, and take actions capable of producing consequences beyond a conversation. But what does it really mean for an agent to learn, cooperate, or deserve our trust?
This book offers an accessible and rigorous introduction to those questions. Drawing on games such as chess, Go, and Mastermind, as well as situations resembling professional work, it distinguishes observable performance from learning, coordination from cooperation, and operational autonomy from the authority to decide.
Ignacio Adrián Lerer examines how instructions, permissions, tools, procedures, and evaluation environments shape what an agent can do. He also explores the problems that arise when multiple agents interact, preserve practices, depend on shared infrastructure, or participate in decisions that affect other people.
The central question is not merely whether these systems work, but under what conditions it is reasonable to trust them. The book examines traceability, reversibility, oversight, authorization, and the distribution of responsibility among those who design, deploy, and use AI agents.
Written for professionals, lawyers, directors, compliance officers, developers, and readers who want to understand agentic artificial intelligence without reducing it to commercial promises or purely technical explanations.