← Develop AI Apps and Agents o… Latent Shelf · Ch.12

Module Study Guide · Path 2 · AI agents

Build Agent-Driven Workflows Using Microsoft Foundry

Define what happens and when — not how — and let the platform orchestrate specialized agents into one coherent process.

— the module's approach

Coordinating multiple agents without code. This module introduces Foundry workflows — a visual, declarative way to orchestrate agents and logic — the workflow patterns for different problems, and the human-in-the-loop controls that keep automation reliable.

visual & declarativenodes & edgeshuman-in-the-loop14 quiz questions
01What workflows are

Orchestration without imperative code

A workflow orchestrates AI-driven actions with a visual, declarative approach: you define a sequence of steps describing what should happen and when, and the platform manages execution and state. Workflows suit business processes that combine AI reasoning, logic, and user interaction.

Key concept · Nodes, edges, and state

A workflow is nodes connected by edges. Each node performs a function — invoking an agent, evaluating a condition, managing data, or communicating with a user — and edges form the execution path. The platform manages state automatically and can pause for human intervention. This is the key differentiator from a single agent: multi-agent workflows combine specialists (classification, decision-making, resolution agents) into a robust, scalable process.

02Patterns

Match the structure to the problem

The workflow structure matters as much as the agents. Choose a pattern by the decision-making, data flow, and human-input the scenario needs.

Workflow patterns.
PatternWhat it isUse for
SequentialFixed step-by-step path; each node passes output to the next.Pipelines, validation, enrichment, response generation.
Human-in-the-loopPauses for user input or approval, then resumes.Approvals, confirmations, filling missing context.
Group chatDynamic orchestration where control shifts between agents by context.Complex, multi-domain requests; collaborative problem-solving.

Key concept · Balancing automation with oversight

Built-in patterns let a workflow pause execution, request human input for critical decisions, and escalate when confidence is low or more context is needed — so real-world systems get efficiency and control.

03Building & maintaining

From canvas to code

You create workflows in Foundry, add agents as nodes, and add logic with Power Fx expressions (the same low-code formula language used across the Power Platform) for conditions and data handling. Workflows can also be maintained and invoked in code for programmatic scenarios and CI/CD. The overall discipline: pick the pattern first, design the node graph, add oversight where risk is highest, then test and iterate.

★Test yourself

Chapter 12 quiz

Fourteen questions on workflows, their components, the patterns, and human oversight. Answer first, then expand Show answer.

Part A · Multiple choice

Q1

A Foundry workflow is best described as:

  • A A single monolithic prompt
  • B A visual, declarative way to orchestrate agents and logic, with the platform managing state
  • C A fine-tuning job
  • D A content filter
Show answer

B. You define what happens and when; the platform handles execution and state.

Q2

What forms the execution path in a workflow?

  • A Edges connecting nodes
  • B A vector index
  • C API keys
  • D Content filters
Show answer

A. Nodes perform functions; edges connect them into the flow.

Q3

Which pattern passes each node's output to the next in a fixed order?

  • A Sequential
  • B Group chat
  • C Human-in-the-loop
  • D Concurrent
Show answer

A — Sequential. Predictable, good for pipelines.

Q4

Which pattern shifts control dynamically between multiple agents by context?

  • A Sequential
  • B Group chat
  • C Human-in-the-loop
  • D None
Show answer

B — Group chat. Agents collaborate and build on each other's outputs.

Q5

Which low-code language adds logic to Foundry workflows?

  • A Power Fx
  • B SQL
  • C Bash
  • D YAML only
Show answer

A — Power Fx. The Power Platform formula language, used for conditions and data handling.

Q6

What advantage do multi-agent workflows have over a single agent?

  • A They can't be paused
  • B They combine specialized agents into a more robust, scalable process
  • C They remove the need for models
  • D They eliminate state
Show answer

B. Classification, decision, and resolution agents form one orchestrated process.

Part B · True or false

Q7

Workflows require writing traditional imperative code to manage execution.

Show answer

False. They're declarative — you describe steps, the platform manages execution and state.

Q8

A workflow can pause to request human approval before continuing.

Show answer

True. Human-in-the-loop is a built-in oversight pattern.

Q9

Workflows can be invoked and maintained in code, not just the visual canvas.

Show answer

True. They support programmatic use for CI/CD scenarios.

Q10

You should choose the workflow pattern after building the whole workflow.

Show answer

False. Pick the pattern based on requirements before designing the workflow.

Part C · Short answer

Q11

Name the two structural elements of a workflow.

Show answer

Nodes (functions) and edges (connections forming the execution path).

Q12

Name the three workflow patterns covered.

Show answer

Sequential, human-in-the-loop, and group chat.

Q13

Give three functions a workflow node can perform.

Show answer

Any three: invoke an agent, evaluate a condition, manage data, or communicate with a user.

Q14

Name three oversight actions built-in patterns support.

Show answer

Pause execution, request human input, and escalate decisions when confidence is low.