Workflow As Agent

What this lesson demonstrates

Both a workflow and an agent expose .run(...), so a whole workflow can be packaged as an agent with workflow.as_agent(name=...) and used anywhere an agent fits — inside another workflow, behind DevUI, or as a tool for a third agent. This lesson wraps a one-node title-casing workflow and calls it exactly like the very first agent lesson — no workflow vocabulary at the call site. Pure executors, zero credentials.

The code

Build an ordinary workflow, then wrap it so it quacks like an agent:

workflow = WorkflowBuilder(start_executor=Headline(id="headline"), name="Headliner").build()
agent = workflow.as_agent(name="HeadlinerAgent")
response = await agent.run("the quick brown fox jumps over the lazy dog")
print(f"agent.run() → {response.text!r}")

The start executor is typed to accept what the wrapper feeds it:

async def run(self, messages: list[Message], ctx: WorkflowContext[Never, str]) -> None:
    text = messages[-1].text  # the latest user turn
    await ctx.yield_output(text.strip().title())

What to notice

• as_agent() gives the workflow an agent interface. The caller uses agent.run("...") with a plain string, exactly as with any agent — the workflow machinery is hidden.
• The wrapper normalizes inputs to list[Message]. A string, a Message, or a list of messages all arrive at the start executor as list[Message].
• Read the latest turn from the list. messages[-1].text pulls the most recent user turn; earlier entries are prior conversation context.

The gotcha

The one hard rule: the workflow’s start executor must accept list[Message] as its input type. The as_agent() wrapper converts every agent-shaped input into list[Message] before handing it to the workflow, so a start node typed for str or int will fail to receive the message.

How it maps to MAF and Foundry

This unifies the two abstractions the series has built separately. Anything that consumes an agent — an orchestration, DevUI hosting, another agent’s tool list — can now consume a full multi-step workflow through the same .run() contract. Composition scales: workflows nest inside agents inside workflows, all speaking one interface, with or without a Foundry model behind any given node.

Run it

uv run tutorial/03-workflows/05_workflow_as_agent.py

Runs offline — no Azure creds. Success: the wrapped workflow title-cases the phrase via .run().

Next: Events