What Makes an AI System Agentic?
The spectrum from chatbot to autonomous agent, where the line is, and what actually crosses it.
The word "agent" gets thrown around a lot. A chatbot is not an agent. A search-augmented chatbot is not really an agent either. An agentic system is one that pursues a goal over multiple steps, takes actions that change the environment it is working in, and adapts its plan based on what actually happens.
Asking a chatbot a question is like asking a contractor for advice. Using an agentic system is like hiring that same contractor to actually build something. They plan the project, order materials, coordinate workers, deal with surprises, and deliver a result. The difference is that they act in the world rather than just respond to it.
Level 1: Tool-augmented chat (single tool call per turn)
Level 2: Orchestrated agent (multi-step, single agent)
Level 3: Multi-agent system (autonomous agents working together)
Level 4: Self-improving agent (modifies its own instructions)
Most production agents today sit at Level 2. Level 3 systems are emerging. Level 4 is largely research territory. Knowing where your system sits on this spectrum helps set the right expectations around reliability, cost, and how much human oversight you actually need.