AI agents

Models that take actions: browsing, running code and using your apps.

What it is

An agent is a model working in a loop: it decides on an action, uses a tool (search, code, a browser, an app), looks at the result and decides what to do next until the task is done.

Why it matters

Agents can finish multi-step jobs, not just answer questions. That's powerful, and it's also why permissions and supervision matter.

Key ideas

• Agent = model + tools + a loop + a goal.
• MCP is an open standard for connecting agents to tools and data.
• Grant the minimum permissions; review before anything irreversible.
• Agents are best on tasks with a clear way to check success.

Try it: Give an agent a checkable job

1. Pick a task with a clear finish line, like “find five papers from 2025 on X and put title, year and DOI in a table.”
2. Run it in an assistant with an agent or research mode.
3. Verify every row. Note where it cut corners.