Skip to content

Challenge 4: Production Workflow

Time: ~20 minutes

Build a multi-agent orchestration workflow for NovaTel Communications and take it to production.

Scenario

The individual agents you built in Challenge 1 are valuable — but in production, agents need to work together as an automated pipeline. In this challenge you wire the two agents into a full call center triage workflow, run it from code, then build and test it visually in the Foundry portal.

deploy

Learning Objectives

  • Deploy persistent production agents (create once, reuse forever)
  • Orchestrate multiple agents step-by-step in a Python workflow
  • Build the same workflow visually in the Foundry portal
  • Invoke the portal workflow from Python with live streaming
  • View run history and traces in the portal

The Workflow

ensure_agents_deployed()
        |
        v
run_intent_classification()     <-- Intent Agent classifies all 7 calls
        |
        v (for each high-priority call)
run_resolution_advisory()       <-- Resolution Agent recommends actions
        |
        v
print_shift_report()            <-- Consolidated Shift Report

Part 1 — SDK: Build and Run the Python Workflow

Step 1: Review the implementation

Open deploy.py and review:

  • ensure_agents_deployed() — lists existing agents, creates intent-classification-agent and resolution-advisor-agent if not present
  • run_intent_classification() — calls the intent agent, handles the lookup_customer function call loop
  • run_resolution_advisory() — calls the resolution agent for each high-priority call
  • run_call_center_workflow() — orchestrates all steps and returns the consolidated report

Step 2: Run the workflow

cd callcenter/challenge-4-deploy
python deploy.py

Expected output: 

=== Step 1: Ensure Agents Are Deployed ===
  Found existing: intent-classification-agent
  Found existing: resolution-advisor-agent

=== Step 2a: Intent Classification ===
  CALL-001: billing_dispute (HIGH) — frustrated, retention risk HIGH
  CALL-002: technical_issue (HIGH) — frustrated, retention risk MEDIUM
  CALL-003: cancellation (HIGH) — neutral, retention risk HIGH
  CALL-004: upsell_opportunity (MEDIUM) — positive, retention risk LOW
  CALL-005: account_support (LOW) — frustrated, retention risk LOW
  CALL-006: billing_dispute (HIGH) — frustrated, retention risk MEDIUM
  CALL-007: security_concern (CRITICAL) — anxious, retention risk MEDIUM

=== Step 2b: Resolution Advisory (High-Priority Calls) ===
  Resolving CALL-007 (security_concern)...
  Resolving CALL-001 (billing_dispute)...
  Resolving CALL-003 (cancellation)...

NOVATEL CALL CENTER — SHIFT REPORT
  Total calls processed  : 7
  Critical priority      : 1
  High priority          : 2
  ...

Part 2 — Portal: Build and Test the Visual Workflow

Step 3: Verify agents are deployed in the portal

  1. Open the Microsoft Foundry portal
  2. Select your project
  3. Left sidebar → Build → Agents
  4. Confirm both agents appear:
  5. intent-classification-agent
  6. resolution-advisor-agent

Step 4: Test the Intent Classification Agent

  1. Click intent-classification-agent → Playground
  2. Send: 
    Classify CALL-001 and CALL-007. What are their intents and priority levels?
  3. Watch the lookup_customer tool call fire for each call
  4. Also try: 
    Classify all calls: CALL-001 through CALL-007. Report intent, priority, sentiment, and retention risk.

Step 5: Test the Resolution Advisor Agent

  1. Left sidebar → Agents → resolution-advisor-agent → Playground
  2. Send: 
    Call CALL-007 from Emma Wilson (basic tier, 8 months):
- Intent: security_concern
- Priority: critical
- Sentiment: anxious
- Context: Received verification codes she didn't request, unfamiliar device on account.
Recommend resolution strategy and immediate actions.
  3. Expected structure: RECOMMENDED ACTION / SCRIPT SUGGESTION / ESCALATION / OFFERS / FOLLOW-UP

Step 6: Build the workflow in the portal designer

  1. Left sidebar → Build → Agents → Workflows → Create → Sequential
  2. In the visual designer:
  3. + Add step → Agent → intent-classification-agent
    • Input: Classify all incoming calls and identify any critical or high-priority items.
  4. + Add step → Agent → resolution-advisor-agent
    • Wire the output of the first step as input to this step
  5. Name it call-center-triage-workflow
  6. Click Save then Deploy

Step 7: Test the workflow in the portal playground

Why you must include the call data in your message

The agents use a lookup_customer tool that reads from a local Python file. The portal playground cannot execute Python functions — if you send a generic prompt, the agent will try to call the tool and stall waiting for a result that never arrives. Paste the call data directly into your message so the agents can work without needing the tool.

  1. Open callcenter-triage-workflow → Playground
  2. Paste the following message (data is pre-embedded so no tool calls are needed):
All call data for today is below — analyse it directly, do not call lookup_customer.

CALL-001 | Maria Gonzalez | premium | 36 months
Unexpected $47.99 charge for sports add-on she never subscribed to. Wants refund, threatening to cancel.

CALL-002 | James Liu | basic | 4 months
Internet dropping every 20-30 minutes since yesterday. Works from home, presentation tomorrow. 1 open ticket.

CALL-003 | Priya Sharma | premium | 18 months
Moving to city without NovaTel coverage, wants to cancel. Asking about ETF and final bill.

CALL-004 | Robert Chen | business | 24 months
Wants to expand from 5 to 12 lines for new hires. Asking about bulk pricing and number porting.

CALL-005 | Sarah Mitchell | basic | 60 months
Confused by new app UI — cannot find billing or data usage pages. 2 open tickets.

CALL-006 | David Park | premium | 12 months
Charged $899 for a device returned 3 weeks ago (has FedEx proof of delivery). 1 open ticket.

CALL-007 | Emma Wilson | basic | 8 months
Suspected account breach — unsolicited SMS verification codes, unfamiliar device on account.

Classify each call by intent, priority, sentiment, and retention risk.
Then recommend resolution strategies for high-priority and security calls.
  1. Watch the steps execute in sequence — classification first, then resolution advisory
  2. Review the final consolidated report

Step 8: Invoke the portal workflow from Python (streaming)

Add to your .env: 

WORKFLOW_AGENT_NAME=call-center-triage-workflow

Re-run the script — Part B activates automatically: 

python deploy.py

You will see each workflow step appear live in the terminal: 

=== Invoking Portal Workflow: call-center-triage-workflow ===
  --> Step: intent_classification
      [completed] intent_classification
  --> Step: resolution_advisory
      [completed] resolution_advisory
<final report streamed here>

Step 9: View run history and traces

  1. Portal → your workflow → Run history tab
  2. Click the latest run to see the execution timeline — each step, duration, and output
  3. Left sidebar → Operate → Tracing to see the full distributed trace across both agent conversations

Success Criteria

  • [ ] Python workflow runs end-to-end: classification → resolution → shift report
  • [ ] Both agents visible in the Foundry portal as persistent assets
  • [ ] Visual workflow created in the portal and tested in its playground
  • [ ] Portal workflow invoked from Python with live step streaming

Beyond the Lab: Production Deployment Options

You've built and tested your agents locally. Here's how to take them to production:

Option 1: Hosted Agents (What You Already Have)

Your agents created with agents.create_version() are already production-ready hosted agents. They live in Foundry indefinitely — any client can invoke them by name via the Responses API. No infrastructure to manage; Foundry handles scaling, versioning, and availability.

  • Versioning: Each create_version() produces an immutable version. Roll back by referencing an older version.
  • Multi-tenant: Multiple users/apps can call the same agent simultaneously.
  • Portal visibility: Agents appear under Build → Agents with playground, run history, and tracing.

Option 2: Foundry Workflows (Visual Orchestration)

What you built in Part 2 — wire multiple hosted agents into a DAG using the portal designer. The workflow becomes a deployable agent invoked via the same Responses API.

  • Step sequencing with automatic output passing
  • Streaming workflow_action events showing progress
  • Run history with per-step timing

Option 3: Azure App Service / Container Apps

Wrap your Python workflow in a FastAPI/Flask app for custom middleware, auth, or business logic:

# Example: FastAPI endpoint that calls your Foundry agents
@app.post("/triage-calls")
async def triage_calls():
    report = run_call_center_workflow(intent_agent, resolution_agent)
    return report

Deploy to App Service (managed PaaS) or Container Apps (auto-scaling containers).

Option 4: Azure Functions (Event-Driven)

Trigger agent workflows from events:

  • Service Bus trigger: Classify and resolve each call as it enters the queue
  • Timer trigger: Generate shift reports every hour during business hours
  • HTTP trigger: On-demand endpoint for supervisors to request triage updates

Pay-per-execution, scales to zero when idle.

Option 5: CI/CD Quality Gates

Integrate evaluation into your deployment pipeline:

  • Run evaluate.py on every PR — block merge if quality drops below threshold
  • Promote agent versions: v1-dev → v1-staging → v1-prod after evaluation passes
  • Blue/green: Deploy new version to 10% traffic, compare metrics, then promote

Summary

Pattern Best For
Hosted Agents Always-on, invoke by name, no infra management
Foundry Workflows Multi-agent orchestration without code
App Service / Containers Custom auth, middleware, webhooks
Azure Functions Event-driven, pay-per-use, queue processing
CI/CD Gates Automated quality assurance before promotion