Developing Azure AI Agents in Azure Functions Using .NET

using Azure.AI.Projects;
using Azure.Identity;
using Microsoft.Azure.Functions.Worker;
using Microsoft.Azure.Functions.Worker.Http;
using System.Net;
public class AskAgent
{
    private readonly AIProjectClient _projectClient;
    public AskAgent()
    {
        _projectClient = new AIProjectClient(
            new Uri(Environment.GetEnvironmentVariable("AI_PROJECT_ENDPOINT")),
            new DefaultAzureCredential());
    }
    [Function("AskAgent")]
    public async Task<HttpResponseData> Run(
        [HttpTrigger(AuthorizationLevel.Function,
        "post")]
        HttpRequestData req)
    {
        var question =
            await new StreamReader(req.Body)
            .ReadToEndAsync();
        var agentClient =
            _projectClient.GetAgentsClient();
        var thread =
            await agentClient.CreateThreadAsync();
        await agentClient.CreateMessageAsync(
            thread.Value.Id,
            MessageRole.User,
            question);
        var run =
            await agentClient.CreateRunAsync(
                thread.Value.Id,
                Environment.GetEnvironmentVariable("AGENT_ID"));
        while (run.Value.Status == RunStatus.Queued ||
              run.Value.Status == RunStatus.InProgress)
        {
            await Task.Delay(1000);
            run = await agentClient.GetRunAsync(
                thread.Value.Id,
                run.Value.Id);
        }
        string responseText = "";
        await foreach (var msg in
            agentClient.GetMessagesAsync(thread.Value.Id))
        {
            if (msg.Role == MessageRole.Agent)
            {
                responseText =
                    msg.ContentItems[0].ToString();
                break;
            }
        }
        var response =
            req.CreateResponse(HttpStatusCode.OK);
        await response.WriteStringAsync(responseText);
        return response;
    }
}

Step 7: Test the Function

Request:

POST https://localhost:7071/api/AskAgent

Body:

What is Generative AI?

Response:

{
  "answer": "Generative AI creates new content such as text, images, code, and other media using AI models."
}

Step 8: Add Custom Tools

Agents become more powerful when connected to business functions.

Example Tool:

public interface IOrderService
{
    Task<OrderStatus> GetOrderStatusAsync(
        string orderId);
}

Example User Request:

Where is my order 12345?

Agent Workflow:

1. Understand the request.
2. Extract order ID.
3. Call GetOrderStatusAsync().
4. Generate a natural language response.
5. Return the result.

Example Response:

Your order 12345 was shipped on 24 October and is expected to arrive tomorrow.

Real-World Use Cases

Customer Support Agent

• Track Orders
• Return Requests
• Customer Queries

HR Assistant

• Leave Balances
• HR Policies
• Employee Information

IT Support Agent

• Password Resets
• Ticket Creation
• Knowledge Base Search

Operations Assistant

• Inventory Lookup
• Shipment Tracking
• Vendor Information

Best Practices

• Use Managed Identity whenever possible.
• Store secrets in Azure Key Vault.
• Keep system instructions focused.
• Log Agent interactions.
• Implement proper error handling.
• Monitor Agent performance using Application Insights.
• Use separate environments for Development and Production.
• Validate user input before sending requests to the Agent.

Common Challenges

Conclusion

Azure AI Agents combined with Azure Functions provide a scalable and serverless architecture for building intelligent applications. By exposing AI Agents through HTTP-triggered Azure Functions, developers can create APIs that understand user intent, execute business operations, and generate natural language responses.

This approach is ideal for customer support systems, enterprise assistants, workflow automation, and modern AI-powered applications.