Agents
AI that doesn’t just answer — it takes steps, uses tools, and works toward a goal without you holding its hand.
Think of it like
Less like a search box, more like an intern you can hand a whole task to.
Example
When a coding assistant reads your files, runs the tests, and fixes what broke, that’s an agent at work.
How it actually works
Under the hood it’s a loop: the model picks an action, something runs it, the result comes back, repeat until done. Most “agent” failures are really loop-control failures — it gets stuck, over-repeats, or wanders off the goal.
For product teams
AI that completes multi-step tasks end to end — automation you delegate outcomes to, not just answers.
For engineers
A model-driven loop: choose action → run tool → observe result → repeat until a stop condition.
Related
- The grown-up version of a chatbot.
- MCP — Only as useful as its tools — which is what MCP standardizes.
- Prompt injection — Its scariest failure mode is prompt injection.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome