What Makes an AI "Agent"?
An AI agent is a system that uses an LLM as its reasoning engine to perceive, decide, and act — not just respond to a single prompt.
The key difference from a chatbot: agency. The system can take multiple steps, use tools, and pursue a goal over time.
The Agent Spectrum
Level 1: Prompt-Response Chatbot
Simple input → LLM → output. No tools, no memory, no planning.
Level 2: Tool-Using Assistant
The LLM can call predefined functions (search, calculator, API calls). ReAct pattern.
Level 3: Multi-Step Planner
Breaks down complex tasks into subtasks, executes them sequentially, handles errors.
Level 4: Autonomous Agent
Sets its own sub-goals, iterates on its approach, uses memory across sessions.
Building Blocks
- LLM: The reasoning core
- Tools: Functions the agent can call
- Memory: Short-term (context window) and long-term (vector DB)
- Planning: Task decomposition and strategy
- Execution: Actually running the planned steps
Common Pitfalls
- Over-engineering: Start at Level 2, not Level 4
- Infinite loops: Always set max iteration limits
- Tool confusion: Keep tool descriptions crystal clear
- Cost blowup: Token usage multiplies fast with multi-step agents