# What is Agentic Workflow?

Also called agent workflow, AI workflow.

An agentic workflow is a defined sequence of work in which one or more AI agents carry out steps, with the surrounding structure specifying the order, the inputs, and the checkpoints. It sits between a rigid automation script and a fully open-ended agent, because the shape of the work is fixed while the content of each step is decided at runtime.

Most production systems land here rather than at either extreme. A developer writes the skeleton, such as gather data, analyze it, draft a response, request approval, then send. Within each stage the agent decides which sources to query and what to write. The skeleton makes the system testable and debuggable, while the agent handles the parts that cannot be enumerated in advance.

Fixing the sequence removes a large class of failures. An agent that plans its own route can loop, skip a required check, or invent a step that was never intended. A workflow cannot, because the transitions are code. The tradeoff is adaptability: work that does not fit the defined shape will fail rather than route around the obstacle.

Common patterns include chaining, where the output of one step becomes the input of the next, routing, where a classifier picks a branch, parallel fan-out with a merge, and iterative refinement, where a step repeats until an evaluator accepts the result. These compose. A complex workflow is usually a handful of these shapes nested inside each other.

A frequent mistake is reaching for a multi-agent design when a workflow would do. Coordination adds cost and failure modes, so the ordinary advice is to use the simplest structure that solves the problem and add autonomy only where the fixed sequence demonstrably fails. Orchestration, task decomposition, and handoffs all describe machinery inside or around workflows.

## Key points

- The sequence is fixed in code; the content of steps is not.
- More predictable and testable than an agent that plans freely.
- Common shapes: chaining, routing, parallel fan-out, refinement loops.
- Fails rather than improvising when work falls outside the shape.

## In practice

An onboarding workflow runs whenever a new customer signs up. Step one pulls company details from a public source. Step two drafts a welcome message referencing what it found. Step three checks the draft against tone rules. Step four holds it for approval. The stages never change, but the research and the wording differ for every customer the workflow touches.

## Related terms

- [Agentic AI](/en/glossary/agentic-ai)
- [Agent Orchestration](/en/glossary/agent-orchestration)
- [Task Decomposition](/en/glossary/task-decomposition)
- [Human in the Loop](/en/glossary/human-in-the-loop)
- [AI Agent](/en/glossary/ai-agent)

[Back to the AI Glossary](/en/glossary)