3.1 — What Are Agents?

The Spectrum

CHATBOT ──────────── ASSISTANT ──────────── AGENT ──────────── AUTONOMOUS SYSTEM
"answers questions"   "does tasks when       "uses tools,         "operates independently
                       asked, one step        makes decisions,      with minimal oversight"
                       at a time"             executes multi-step
                                              workflows"

Chatbot — responds to questions. No memory between sessions, no tools, no actions. Pure text in, text out.

Assistant — can do things when asked. Follows instructions step by step. Needs direction for each action.

Agent — has tools, makes decisions about which tools to use, executes multi-step plans. Can read files, search the web, write code, run commands. Has a degree of autonomy within guardrails.

Autonomous System — operates continuously with minimal human oversight. Monitors, decides, acts. Requires serious guardrails and human-in-the-loop mechanisms for critical decisions.

Claude Code Is an Agent

You give it a goal (“fix this bug”), it decides which files to read, what the problem is, how to fix it, tests the fix, and reports back. That’s agency — not just Q&A.

The difference is not how smart the underlying model is. It’s what the system is allowed to do. A chatbot and an agent can share the same model — the agent just has keys to more rooms.

Tool Use

Tool use / Function calling — the mechanism that lets an AI do things beyond generating text. When Claude Code reads a file, runs a command, or searches code — it’s using tools. The AI doesn’t “know” how to read files — it has access to a tool that reads files, and it decides when to use it.

Analogy: A new employee who can answer questions (chatbot) vs. one who also has keys to the filing cabinet, access to the database, and authority to send emails (agent). Same brain, different access.

Tools are what make agents agents. Without tools, an AI can only generate text. With tools, it can take actions in the world — read your files, call APIs, write to a database, trigger deployments.

This is the foundation of everything else in Phase 3: once an AI has tools, you need to think carefully about what tools it has, what rules it follows, what it remembers, and when a human needs to stay in the loop.

---

Next: 3.2 — The Agent Stack | Phase overview: Phase 3