WTF is Agentic Workflow?
LLM calls wired together along a path you define in code, rather than letting the model choose every step.
Also called: workflow, prompt chaining, llm workflow
TL;DR
Not every AI feature needs a fully autonomous agent. A workflow is a fixed recipe: step 1 summarise, step 2 classify, step 3 draft a reply. It is more predictable and cheaper than an open-ended agent, and often good enough.
How it works
Anthropic's taxonomy distinguishes workflows (predefined code paths) from agents (the model directs its own process). Common workflow patterns: prompt chaining (sequential steps with checks between them), routing (classify then dispatch to a specialised prompt), parallelisation (sectioning or voting), orchestrator–workers, and evaluator–optimizer. Rule of thumb: start with the simplest workflow that works and add autonomy only when the task truly needs it.
Origin
- Coined / introduced by
- Popularised by Andrew Ng; formalised by Anthropic
- When
- 2024
- Story
- 'Agentic workflows' spread in early 2024; Anthropic's 'Building effective agents' (Dec 2024) gave the widely used workflow-vs-agent distinction.