An AI agent combines 3 LLM capabilities:
- Reasoning: breaks down a goal into steps
- Tools: calls external functions (browser, API, code, files)
- Memory: keeps context across multiple steps
Examples 2026: Claude Computer Use, OpenAI Operator, Devin, AutoGPT.
Why it matters: shift from "answering tool" to "task executor". Multiplied productivity, but new risks (autonomy = loss of control).