Agentic Workflow
A system design pattern in which the model dynamically decides at runtime the next action, which tool to invoke, and whether follow-up questions are needed, as distinguished from traditional workflow automation in which humans pre-orchestrate each step.
Timeline (last 90 days)
Traditional automation is predictable and easy to audit but struggles with ambiguous inputs; agentic workflows are flexible but introduce non-determinism and new failure modes (such as misjudging intent, loops, and excessive calls).
When only an explanation or draft is needed, a single conversation may suffice; agentic workflows truly show their value when execution spans multiple tools and multiple steps.
In practice, deterministic processes and agentic workflows are often mixed: deterministic processes serve as the skeleton, and at nodes requiring judgment, decisions are handed to the model.
All Facts (3)
Traditional automation is predictable and easy to audit but struggles with ambiguous inputs; agentic workflows are flexible but introduce non-determinism and new failure modes (such as misjudging intent, loops, and excessive calls).
50%UnverifiedWhen only an explanation or draft is needed, a single conversation may suffice; agentic workflows truly show their value when execution spans multiple tools and multiple steps.
50%UnverifiedIn practice, deterministic processes and agentic workflows are often mixed: deterministic processes serve as the skeleton, and at nodes requiring judgment, decisions are handed to the model.
50%