An AI agent doesn't just generate text — it uses tools, browses the web, writes and runs code, and takes multi-step actions to complete goals autonomously.

An **AI agent** is an LLM equipped with tools and the ability to plan multi-step actions.

**Core loop (ReAct):**

1. **Observe** the environment / task

2. **Think** about what to do next

3. **Act** using a tool (search, code interpreter, API call)

4. **Observe** the result

5. Repeat until task complete

**Common tools:** web search, code execution, file read/write, API calls, browser control

**Frameworks:** LangChain, LlamaIndex, Autogen, Claude's tool use, OpenAI Assistants

**Real-world use:** customer support, data analysis, software engineering (Devin, Cursor), research assistants.

AI Agents: When Models Can Take Actions