Agentic AI
Agentic AI
Also called: 智能体式 AI · Agentic system
An umbrella label for AI systems that emphasise autonomous decision-making and multi-step execution; it names a design orientation, not a model type.
What the word points to
"Agentic" is a term of degree: a system is more agentic the more it can set its own sub-goals, choose actions, and adjust based on results. The same underlying model can be used for one-shot Q&A or wrapped into a highly autonomous agent.
So "agentic AI" usually describes a product and engineering direction — how model, tools, memory, and control flow are assembled — rather than a new piece of technology.
Why it surged in 2025–2026
As tool calling, long context, and reasoning models matured, running a model for dozens of steps without derailing became practical, and the industry narrative moved from "chat assistant" to "agent that does work." The term is now also heavily used in marketing, so it is worth separating capability claims from verified deployment results.
Common misconceptions
- It is not a model category — you cannot "download an agentic AI."
- "More agentic" is not always better: higher autonomy needs more guardrails, observability, and cost control.
Related terms
Sources
Compiled 2026-08-29 · This glossary is compiled from public papers, official specifications, and common industry definitions, and is updated as the field evolves. Corrections welcome.