concept · weight 8
AI Agents
LLM-driven systems that plan, call tools, and act toward a goal.
An AI agent wraps a language model in a loop: it observes a goal, decides on an action (often calling a tool), runs that action, observes the result, and repeats until done. Coding agents like Claude Code are a concrete example.
The agent loop
| Step | What happens |
|---|---|
| Perceive | Read the task, context, and prior results |
| Decide | The model chooses an action or a tool call |
| Act | Run the tool (shell, search, file edit, API call) |
| Observe | Feed the result back into context |
| Repeat | Continue until the goal is met or a stop condition fires |
Building blocks
- Tools — typed functions the model can invoke, usually over an API.
- Memory — short-term context plus optional long-term storage.
- Planning — breaking a goal into steps, sometimes with self-critique.
- Guardrails — permissions and validation around risky actions.
The quality of an agent depends less on the model alone and more on the tools, context, and feedback loop it operates within.