Part 6 of 9 · 1 min read
What agentic AI means
The field, and why the word is about degree, not a yes-or-no label.
"Agentic" is an adjective. It describes how much a system acts on its own to reach a goal.
Agentic AI
Agentic AI is software where a language model takes steps toward a goal on its own: it chooses actions, uses tools, looks at the results and decides what to do next. It always comes with limits and checks that keep it safe and useful. The field of agentic AI is about building such systems well.
Example: a support system that reads a customer's message, looks up their order, decides whether a refund is allowed, issues it, and writes the reply, while stopping to ask a person before refunds above a set amount.
"Agentic" is a matter of degree. A chatbot is not agentic at all. A workflow with one flexible step is a little agentic. A system that plans, uses many tools and works for minutes on its own is very agentic.
Autonomy
Autonomy is how much the system decides without a person. Low autonomy means a person approves every step. High autonomy means it runs on its own and a person only reviews the result.
Example: a draft email that you read and send yourself is low autonomy. An assistant that sends emails on its own is high autonomy.
More autonomy is not always better. It saves time, but each decision made without a person is a decision nobody checked. Good agent builders pick the lowest autonomy that still does the job.
Check yourself:
- Is a system that always runs the same three steps "agentic"? Why or why not?