What an agent is
An LLM agent wraps a language model in a loop that lets it take actions, observe results, and decide what to do next. The model alone only predicts text. The architecture around it gives it agency.
The standard components
- Model the reasoning engine that picks the next action
- Tools functions the model can call to read or change the world
- Memory state carried across steps so the agent does not forget
- Orchestrator the loop that runs the model, executes tools, and feeds results back
The control loop
The agent reads a goal, thinks, optionally calls a tool, sees the observation, and repeats until it decides it is done. Each turn appends to a growing context.
Good architectures keep this loop observable and bounded so it cannot run forever. A step budget and clear stop conditions matter as much as the model itself.
Key idea
An agent is a model plus a loop: reason, act, observe, repeat, until a goal is met or a budget is spent.