References: What makes an AI an "agent"

Source material

Source curriculum (structural mirror, cited as further study):
• Microsoft, "Introduction to AI Agents and Agent Use Cases"
  (AI Agents for Beginners, Lesson 01)
  Author: Microsoft Cloud Advocates
  Lesson page: https://github.com/microsoft/ai-agents-for-beginners/tree/main/01-intro-to-ai-agents
  License: MIT
• Berkeley CS294/194-196, "LLM agents: brief history and overview" (Lecture 2)
  Speaker: Shunyu Yao
  Course page: https://rdi.berkeley.edu/llm-agents/f24
  Video: https://www.youtube.com/watch?v=RM6ZArd2nVc
  License: YouTube standard (link-out only)
Clawdemy's lessons are original prose that follows the pedagogical arc of these
sources. We do not reproduce or transcribe them; we cite them as recommended
companions. All rights to the original materials remain with their authors.

Read or watch this next

Introduction to AI Agents and Agent Use Cases (Microsoft) by Microsoft Cloud Advocates. The practitioner version of this lesson, with runnable code samples and a section on when to use agents. MIT-licensed, free to read and fork. Start here if you want to see the loop in working code rather than pseudocode.
LLM agents: brief history and overview (Shunyu Yao) from Berkeley’s CS294 LLM Agents course. The academic framing for the history section of this lesson: where agents came from and how the language-model era changed the “decide” step. About an hour, no math prerequisites.

Going deeper on the history

A short, durable list. Each is a primary source for a named result in the lesson.

ELIZA, A Computer Program For the Study of Natural Language Communication Between Man and Machine (Weizenbaum, 1966). The original paper for the 1960s rule-based conversational program. The starting point for “agents predate language models.”
STRIPS: A new approach to the application of theorem proving to problem solving (Fikes & Nilsson, 1971). The symbolic planner that searched for action sequences to reach a goal. The “explicit search” decision style the lesson contrasts with language-model decisions.
A robust layered control system for a mobile robot (Brooks, 1986). The subsumption architecture: reactive robotics that wired perception directly to action. The 1980s reactive-agent reference.

Adjacent topics

Where this leads inside this track.

How tool use turns a model into an agent. The next lesson. It opens the loop and traces a single tool call in detail: how the model emits a structured request, how the outside loop runs it, and how the result comes back. The mechanism behind every example here.
The tool-use design pattern in depth. A few lessons ahead. Once you have seen one tool call, the question becomes how to define a tool so the model reliably knows when and how to use it. That is a design pattern with its own pitfalls.
Building trustworthy agents and securing agents. Later in the track. A loop that can act in the world can also act wrongly or be manipulated. Those lessons cover the failure modes and the attack surface that come with agency.