Agentic workflow
What it means
What is an agentic workflow?
An agentic workflow is a task one or more AI agents complete on their own once started. The system plans steps, calls tools or APIs, checks the result, and decides what happens next without a person approving each move.
The distinction that matters is between a single prompt and response and a task an agent sees through to the end. A chatbot that answers one question is not agentic. A system that reads a support ticket, looks up the order, drafts a refund and issues it is agentic. Nobody scripted that exact sequence in advance.
Engineering teams build this pattern because it automates the cross-system judgement calls that used to need a person. Security reviewers and boards ask about it separately from the model itself, because the risk changes once a system can act, not only speak.
A wrong sentence in a chat reply is embarrassing. A wrong action against a live order or a payment has already happened by the time anyone notices, because nothing stood between the decision and the outcome.
Published · Updated
In practice
Why it comes up
It comes up the moment a pitch or a product description says “agent” instead of “assistant”. A security reviewer’s next question is almost always what the agent can do without asking first. The reviewer asks because that is where the real exposure sits, not the model choice. A board hears “agentic” and correctly hears “this thing takes actions on its own.” That is a different risk conversation than a chatbot that only produces text.
What to do about it
List every action an agent can take, not just what it can say, and sign off on each as a separate risk. Set a real limit on what it can do without a person confirming first, especially anything that spends money, touches a customer record, or sends something external. Log every tool call, so a bad outcome traces to a specific decision. Evals belong here: test the whole action chain, not just the model’s replies.
Sources and neighbours
Sources
Related terms