05 · First Workflow

What this lesson demonstrates

The first four lessons were about a single agent calling a model. A Workflow is the framework’s other primitive: a directed graph of executors (processing units) connected by edges (data-flow paths). No model is involved at all — each executor is a plain Go function. This is the smallest possible workflow: two executors in a line. For input "Hello, World!" the pipeline uppercases to "HELLO, WORLD!" then reverses it, producing "!DLROW ,OLLEH".

The code

Each executor wraps a plain function; the builder wires them and picks the output node:

uppercase := workflow.NewExecutor("UppercaseExecutor", func(input string) string {
    return strings.ToUpper(input)
}).Bind()

reverse := workflow.NewExecutor("ReverseExecutor", func(input string) string {
    runes := []rune(input)
    slices.Reverse(runes)
    return string(runes)
}).Bind()

return workflow.NewBuilder(uppercase).
    AddEdge(uppercase, reverse).
    WithOutputFrom(reverse).
    Build()

What to notice

• Executors wrap plain functions. workflow.NewExecutor(name, func(string) string {…}) turns an ordinary Go func into a graph node; .Bind() produces an ExecutorBinding the builder can reference in edges.
• The builder is fluent and validated. NewBuilder(uppercase) sets the start node, AddEdge(uppercase, reverse) routes one’s output into the other, WithOutputFrom(reverse) names the output node, and Build() validates the graph before returning a *Workflow. Forgetting WithOutputFrom is the easy gotcha — the graph runs but yields nothing.
• Execution is event-driven. inproc.Default.Run(ctx, wf, input) runs the graph in-process on a background goroutine and returns a *Run. Each executor that finishes emits an ExecutorCompletedEvent (ExecutorID + Result); ranging run.NewEvents() lets you watch data flow through the pipeline.

How it maps to Azure AI Foundry

Nothing here touches Foundry — and that’s the point. Workflows and agents are orthogonal primitives. An executor can call an agent (and later lessons put agents inside workflows), but the workflow engine itself is pure orchestration: build a graph, run it, iterate its events. Learning the graph mechanics with plain string functions first means the model calls later drop into a structure you already understand.

Run it

go run ./tutorial/01-get-started/05_first_workflow

Expected output is the input followed by one line per executor as it completes, ending in ReverseExecutor: !DLROW ,OLLEH. This lesson is fully offline: main_test.go builds the same graph, runs it, and asserts the final output — no model, no credential, no AF_LIVE gate needed.

Next: a2a · Remote Skills as Function Tools