What is the difference between agentic AI and generative AI?

Generative AI produces content – text, images, code, or summaries – in response to an input; a person decides what happens with it. Agentic AI uses generative models as a reasoning tool, but pursues a goal across multiple steps with them, calls systems, and carries out actions. The difference lies in the outcome: with generative AI it is content, with agentic AI it is a changed state in the world.

Why does this matter?

The two terms are often mixed up, yet they require different safeguards. With generative AI a quality check of the output is usually enough; with agentic AI it must be defined which systems the agent may change, when it needs approval, and how its actions are logged.

What does this look like in practice?

Asked "why did delivery times rise in March?", a generative AI writes a plausible explanation. An agentic system queries the process model, compares March with February, finds a warehouse where picking has stalled since a shift change, opens a ticket for the site manager, and checks a week later whether the waiting time has come down.

Not to be confused with

Not an opposition: agentic AI builds on generative AI rather than replacing it. Nor the same as the copilot-versus-agent distinction, which concerns how the system works with a person; this one concerns what kind of result it produces.