An AI agent is any application component where LLM outputs influence workflow decisions or actions. In practice, this spans from:
- Basic LLM function calling and tool use
- Predefined sequences of LLM-driven actions
- Single-purpose AI services
- Autonomous decision-making components
- Multistep task planning and execution
- Complex service coordination
Think of agents as building blocks in your AI application - whether they're making decisions about customer support queries, generating content, or coordinating with other services. What matters isn't the label "agent," but rather how these LLM-powered components work together reliably in production.
For more on this, see Anthropic's What are agents?.