Agents & Tools
Theory
Chat-only AI talks. An agent acts.
The shift comes from three pieces working together:
- a planner — the model breaks a goal into smaller steps
- tools — functions the model can call to read or change the world (function calling, often via MCP)
- a loop — the model checks the result and decides what to do next
- Plan: "To answer this, I need today's weather and your calendar."
- Act: call the weather tool, call the calendar tool.
- Observe: "Rain at 3pm; you are free at 5pm."
- Decide: "Suggest the 5pm slot, ask for confirmation."
Common everyday agents you can already use:
- email triage (label, summarize, draft replies — you send)
- calendar assistant (suggest slots — you confirm)
- research helper (fetch, summarize, cite — you decide)
A good rule of thumb: an agent should never do a thing you would not let an unfamiliar coworker do on your behalf.