Skip to main content

Challenge 22: Azure AI Agent Service

Estimated Time

60 min | Cost: $3-8 (estimated) | Domain: Implement Agentic Solutions (5-10%)

Exam skills covered​

  • Create an agent with Azure AI Agent Service
  • Configure agent tools (file search, code interpreter)
  • Manage threads and runs for agent conversations

Overview​

Azure AI Agent Service provides a managed platform for building AI agents. It handles thread management, tool execution (file search, code interpreter, custom functions), run lifecycle, and file management.

The architecture: create an agent → create a thread → add messages → create a run → poll until complete → read responses.

Prerequisites​

  • Azure subscription
  • Azure AI Foundry hub and project
  • Python 3.9+ or .NET 8
  • Packages: azure-ai-projects, azure-identity

Implementation​

Task 1: Set Up Azure AI Foundry Project​

az group create --name rg-ai102-foundry-agents --location eastus2

az extension add --name ml

az ml workspace create \
  --name hub-ai102-agents \
  --resource-group rg-ai102-foundry-agents \
  --kind hub \
  --location eastus2
import os
import time
from azure.ai.projects import AIProjectClient
from azure.ai.projects.models import FilePurpose, FileSearchTool, RunStatus
from azure.identity import DefaultAzureCredential

project_client = AIProjectClient(
    credential=DefaultAzureCredential(),
    endpoint=os.environ["AZURE_AI_PROJECT_ENDPOINT"],
)

# Upload a knowledge file
with open("pricing-info.md", "w") as f:
    f.write("# Azure AI Pricing\n- GPT-4o: $2.50/1M input, $10/1M output\n- Vision: $1/1K calls\n")

uploaded_file = project_client.agents.upload_file(
    file_path="pricing-info.md",
    purpose=FilePurpose.AGENTS
)

vector_store = project_client.agents.create_vector_store_and_poll(
    file_ids=[uploaded_file.id],
    name="pricing-kb"
)

# Create agent with file search tool
agent = project_client.agents.create_agent(
    model="gpt-4o",
    name="Pricing Assistant",
    instructions="You are a pricing assistant. Use file search to answer questions accurately.",
    tools=[FileSearchTool()],
    tool_resources={"file_search": {"vector_store_ids": [vector_store.id]}}
)
print(f"Created agent: {agent.id}")

# Create thread and run
thread = project_client.agents.create_thread()
project_client.agents.create_message(
    thread_id=thread.id, role="user",
    content="How much does GPT-4o cost per million input tokens?"
)

run = project_client.agents.create_run(thread_id=thread.id, agent_id=agent.id)

while run.status in [RunStatus.QUEUED, RunStatus.IN_PROGRESS]:
    time.sleep(1)
    run = project_client.agents.get_run(thread_id=thread.id, run_id=run.id)

if run.status == RunStatus.COMPLETED:
    messages = project_client.agents.list_messages(thread_id=thread.id)
    for msg in reversed(messages.data):
        if msg.role == "assistant":
            for block in msg.content:
                if hasattr(block, "text"):
                    print(f"Assistant: {block.text.value}")

project_client.agents.delete_agent(agent.id)

Task 3: Code Interpreter Tool​

from azure.ai.projects.models import CodeInterpreterTool

code_agent = project_client.agents.create_agent(
    model="gpt-4o",
    name="Data Analyst",
    instructions="You are a data analyst. Use code interpreter for calculations and charts.",
    tools=[CodeInterpreterTool()]
)

thread = project_client.agents.create_thread()
project_client.agents.create_message(
    thread_id=thread.id, role="user",
    content="Calculate monthly cost: 500K GPT-4o input tokens/day, 100K output/day. Show breakdown."
)

run = project_client.agents.create_and_process_run(
    thread_id=thread.id, agent_id=code_agent.id
)

if run.status == RunStatus.COMPLETED:
    messages = project_client.agents.list_messages(thread_id=thread.id)
    for msg in messages.data:
        if msg.role == "assistant":
            for block in msg.content:
                if hasattr(block, "text"):
                    print(block.text.value)

project_client.agents.delete_agent(code_agent.id)

Expected Output​

Challenge 22 - Multi-Agent Conversation Flow

Break & fix​

ScenarioSymptomRoot CauseFix
Run stuck in in_progressTimeoutLarge file indexing or model latencyUse create_and_process_run with timeout; check vector store status
requires_action statusRun pausesAgent needs function call resultsSubmit tool outputs via submit_tool_outputs
File search returns nothingGeneric responseFile not indexed yetWait for vector store completed status before running
404 on threadThread not foundThread ID incorrect or deletedVerify thread ID; threads persist until explicitly deleted
Model deployment errorAgent creation failsModel not deployed in projectDeploy model via AI Foundry portal first

Knowledge Check​

1. What is the correct sequence for using Azure AI Agent Service?

2. What does the file_search tool provide in Azure AI Agent Service?

3. What is the purpose of the code_interpreter tool?

4. How should you handle the 'requires_action' run status?

5. What is the relationship between agents and threads?

Cleanup​

az group delete --name rg-ai102-foundry-agents --yes --no-wait

Learn More​