Functional Workflow Basics

What this lesson demonstrates

To see the workflow mechanics without the LLM in the way, this program strips out the agents entirely. It classifies a string as “short” or “long” using two @step functions composed by a @workflow. No model, no az login, no network — just the orchestration machinery, isolated.

The code

A @step is an async unit of work whose result is cached and checkpointed; a @workflow composes steps with ordinary control flow:

@step
async def normalize(text: str) -> str:
    return text.strip().lower()

@step
async def word_count(text: str) -> int:
    return len(text.split())

@workflow
async def summarize_input(raw: str) -> str:
    clean = await normalize(raw)
    count = await word_count(clean)
    verdict = "short" if count < 5 else "long"
    return f"{count} words ({verdict}): {clean!r}"

What to notice

• Plain Python is the orchestration. Sequential awaits and an ordinary if branch — no DSL to learn before the graph API arrives in the next lesson.
• @step is where checkpointing lives. Each step’s result is cached/checkpointed, which is what makes workflows resumable later; here it just runs top to bottom.
• The gotcha: .run() returns a WorkflowRunResult, so read the value with result.get_outputs()[0] — get_outputs() returns a list because a workflow can yield more than one output.

How it maps to Azure AI Foundry

Nothing — deliberately. This lesson has no FoundryChatClient and makes no Foundry call, which is the point: it isolates the workflow engine from the model so you can reason about steps, caching, and outputs on their own before adding agents back in.

Run it

uv run tutorial/01-get-started/06_functional_workflow_basics.py

Runs offline — no Azure creds needed. Expected: one line classifying a 5-word string as “long”.

Next: First Graph Workflow