AI agent
By HubAI editorial team · Source checked: 24 October 2026What does it mean?
A system that uses a model to choose actions towards a goal, often by calling tools and checking results over several steps. Products use the term loosely; inspect the actual behaviour.
Picture this
You ask a system to prepare a meeting pack. It finds authorised documents, drafts a briefing and asks before sending it. The permission and approval steps are part of the design.
Where you’ll come across it
Multi-step research, coding and administrative workflows. Start with a narrow, reversible task and a clear stopping condition.
What it doesn’t mean
An agent is not a trustworthy employee by default. Limit access, cap time and spending, and keep an audit trail of actions rather than only the final answer.
Draw the permission boundary
A short exercise · use fictional or non-sensitive material.
- Sketch a simple assistant that drafts a reply but cannot send it. Use invented recipients.
- Mark which tools it can read from, which it could write to and where approval is required.
- Try an instruction inside a sample document asking it to ignore those limits. Check that actual permissions still block the action.
Which safeguard is enforced by the application rather than a polite instruction?
Did that make it clearer?
Likes are a private on-page choice, not a public rating. For a correction or complaint, send the pre-filled form to our editorial inbox. Nothing is sent until you submit it.Check the source
Written and explained by HubAI. The reference below lets you go deeper; it is not an endorsement of a particular product.
Technical reference: huggingface.co ↗