Agentic AI

Agentic AI describes AI systems that pursue a goal over multiple steps, choosing actions and calling tools along the way, rather than returning a single response to a single prompt.

The practical difference from an assistant is autonomy over sequence. You state an outcome, and the system decides which steps to take and which tools to call to get there, adjusting as results come back.

That capability changes the failure mode. A chatbot's mistake is a wrong answer a person can ignore. An agent's mistake is a sequence of real actions taken against real systems, each one plausible, discovered later.

Which is why agentic deployments tend to succeed or fail on boundaries rather than on model quality: a scoped identity, an approved tool list, a stopping point before anything irreversible, and a record of every run.