Hands-On RAG for Production

Chapter 7. From RAG to AI Agents

The definition of a software agent first emerged from the fields of distributed AI and computer science in the 1970s and 1980s. A foundational concept was Carl Hewitt’s Actor Model, first proposed in 1973, which defined self-contained, interactive components (“Actors”) that communicate through asynchronous message passing.

While not explicitly called “agents” at first, Actors embodied the core principles of autonomy and concurrent operation. By the 1980s, researchers used the term “agent” to describe a software entity with specific characteristics: it was autonomous (in control of its own actions), had “social” abilities (could interact with other agents), was reactive (responded to its environment), and was proactive (took initiative to meet its goals). These early agents emerged from the field of symbolic AI, operating on “hand-coded” rules and logical inference within highly structured, limited environments.

The 1990s became the “age of intelligent agents,” where these concepts were formalized and put into practice. A key development was the Belief-Desire-Intention (BDI) model, an architecture that gave agents a more human-like reasoning structure. An agent would maintain beliefs about the state of the world, have desires (goals to achieve), and form intentions (committed plans of action). This era saw the rise of early practical agents, such as information agents that could filter emails or search nascent databases, and the first multi-agent ...