Glossary
What is an agentic workflow?
An agentic workflow is a process where language models carry out several steps, often with tools, to complete a task. Some follow fixed steps written in code (workflows); others let the model choose steps (agents). Common patterns include chaining prompts, routing, running in parallel, and evaluating and refining results.
Common patterns
| Pattern | What it does |
|---|---|
| Prompt chaining | Each step's output feeds the next |
| Routing | Classify input, send to the right handler |
| Parallelisation | Run subtasks at once, combine results |
| Orchestrator-workers | One model splits work for others |
| Evaluator-optimiser | One model drafts, another critiques, repeat |
These are described in Anthropic's "Building effective agents", which advises starting with the simplest approach and adding autonomy only when needed.
Workflow or agent?
If the steps are known and stable, a fixed workflow is cheaper and more reliable. If each case differs, like submit forms that vary by site, an AI agent adapts better. Many products combine both: a fixed outer process with an agent handling the variable part.
Example
A launch process runs as a workflow: prepare copy, pick directories, then for each directory hand the form to an agent, then check results and update a tracker. The agent part handles each site's quirks; the workflow keeps the whole thing predictable.
FAQ
Is an agentic workflow the same as an AI agent?
Related. Workflows follow coded steps; agents choose steps. Both are agentic systems.
Do I need a framework?
Not necessarily; many workflows are a few API calls in plain code.
When should I add more autonomy?
When fixed steps can't handle the variety of real cases.