AI docs · Building with AI
AI agents
Systems where an LLM plans and takes actions through tools in a loop, rather than producing a single response.
What it is
- An agent uses an LLM to decide what to do, call tools (search, code, APIs), observe results, and continue until a goal is met.
- It turns a model from a text generator into something that can act on the world.
How it works
- The model is given a goal, a set of tools, and guardrails.
- It reasons, calls a tool, reads the result, and repeats, maintaining state across steps.
- Good agents constrain the loop with limits, validation, and human checkpoints for risky actions.
Trade-offs
- Can automate multi-step work, but reliability drops as the number of steps grows.
- More autonomy means more risk: errors and tool misuse compound.
When to use it
- Multi-step tasks that need tools and adaptation, where some unreliability is acceptable or checked.
- When a fixed workflow is too rigid but full autonomy is unnecessary.
Common pitfalls
- Giving agents powerful tools without guardrails or human approval.
- Long, open-ended loops that drift, loop, or rack up cost.
Quick check
What most distinguishes an AI agent from a single model call?
Related concepts
Let's get to work
Want to build with AI for real?
Beyond the explainer, we design, secure, build and run production AI. Tell us what you have in mind.