Build an Agentic AI System in .NET 8 Using OpenAI and C#

Introduction

Most of us have already used Generative AI tools that respond to prompts — they write code, summarize text, or generate SQL when asked.

But the next evolution is Agentic AI.

Agentic AI doesn’t just respond — it plans, decides, and executes tasks to achieve a goal.

Instead of:

• “Tell me what’s wrong.”

You assign:

• “Investigate failures and fix them.”

The AI becomes a digital worker embedded inside your system.

This article walks through building a real-world Agentic AI Proof of Concept using:

• .NET 8 Web API

• SQL Server

• Semantic Kernel

• Background Services

• OpenAI (LLM)

We will create an Autonomous Order Reconciliation Agent that detects and fixes failed orders automatically.

Solution Architecture

Step 1 — Basic Setup (.NET + SQL Server)

Create a new Web API project:

• dotnet new webapi -n OrderAgentPOC

• cd OrderAgentPOC

Install required packages:

• dotnet add package Dapper

• dotnet add package Microsoft.Data.SqlClient

• dotnet add package Serilog.AspNetCore

Configure Database Connection

appsettings.json

"ConnectionStrings": { 
   "DefaultConnection": "Server=localhost;Database=OrderAgentPOC;Trusted_Connection=True;TrustServerCertificate=True;" 
}

Create Database

CREATE DATABASE OrderAgentPOC; 
GO CREATE TABLE Orders 
( 
Id INT IDENTITY PRIMARY KEY, 
OrderNumber NVARCHAR(50), 
Amount DECIMAL(10,2), 
Tax DECIMAL(10,2) NULL, 
Status NVARCHAR(50), 
CreatedDate DATETIME DEFAULT GETDATE() 
); 

INSERT INTO Orders (OrderNumber, Amount, Tax, Status) 
VALUES 
('ORD-1001', 1000, NULL, 'Failed'), 
('ORD-1002', 500, 50, 'Completed'), 
('ORD-1003', 750, NULL, 'Failed');

These failed rows simulate real reconciliation issues.

Create DB Connection Factory

public class DbConnectionFactory 
{ 
  private readonly IConfiguration _configuration; 
  public DbConnectionFactory(IConfiguration configuration) 
  { 
     _configuration = configuration; 
  } 
  public IDbConnection CreateConnection() 
    => new SqlConnection(_configuration.GetConnectionString("DefaultConnection")); 
}

Step 2 — Tools + DB Simulation (Expose System Actions to AI)

Agentic AI works through Tools.

Think of tools as:

Functions the AI can call to interact with your system.

Create AI Tools

public class OrderAgentTools 
{ 
    private readonly DbConnectionFactory _factory; 
    public OrderAgentTools(DbConnectionFactory factory) 
    { 
       _factory = factory; 
    } 

    [KernelFunction, Description("Get failed orders")] 
    public async Task<string> GetFailedOrders() 
    { 
      using var conn = _factory.CreateConnection(); 
      var orders = await conn.QueryAsync( "SELECT Id, OrderNumber, Amount, Tax FROM Orders WHERE Status='Failed'"); 
      return JsonSerializer.Serialize(orders); 
    } 

    [KernelFunction, Description("Fix missing tax")] 
    public async Task<string> FixMissingTax(int orderId) 
    { 
      using var conn = _factory.CreateConnection(); 
      await conn.ExecuteAsync(@" UPDATE Orders SET Tax = Amount * 0.18, Status = 'Completed' WHERE Id = @Id", 
        new { Id = orderId }); 
      return $"Order {orderId} fixed."; 
     } 
}

These methods are now AI-callable actions.

Step 3 — Agent Integration (Add Reasoning)

Install Semantic Kernel:

• dotnet add package Microsoft.SemanticKernel

• dotnet add package Microsoft.SemanticKernel.Connectors.OpenAI

Configure Kernel (Scoped Lifetime)

builder.Services.AddScoped<Kernel>(sp => 
{ 
  var config = sp.GetRequiredService<IConfiguration>(); 
  var builder = Kernel.CreateBuilder(); 
  builder.AddOpenAIChatCompletion( modelId: config["OpenAI:Model"], apiKey: config["OpenAI:ApiKey"] ); 
  return builder.Build(); 
});

Create the Agent

public class ReconciliationAgent 
{ 
  private readonly Kernel _kernel; 
  public ReconciliationAgent(Kernel kernel, OrderAgentTools tools) 
  { 
    _kernel = kernel; _kernel.ImportPluginFromObject(tools, "OrderTools"); 
  } 
  public async Task<string> RunAsync() 
  { 
     var prompt = @" 
You are an autonomous reconciliation agent. 

1. Fetch failed orders. 
2. Identify missing tax. 
3. Fix them using tools. 
4. Confirm completion."; 
    var result = await _kernel.InvokePromptAsync(prompt); return result.ToString(); 
  } 
}

Now the AI decides when to call which tool.

Step 4 — Background Automation (Make It Autonomous)

This converts the solution from AI-assisted to Agentic

Background Worker

public class AutonomousAgentService : BackgroundService 
{ 
   private readonly IServiceScopeFactory _scopeFactory; 
   private readonly ILogger<AutonomousAgentService> _logger; 
   public AutonomousAgentService(IServiceScopeFactory scopeFactory, ILogger<AutonomousAgentService> logger) 
   { 
     _scopeFactory = scopeFactory; 
     _logger = logger; 
   } 
   protected override async Task ExecuteAsync(CancellationToken stoppingToken) 
   { 
     _logger.LogInformation("Autonomous Agent Started"); 
     while (!stoppingToken.IsCancellationRequested) 
     { using var scope = _scopeFactory.CreateScope(); 
       var agent = scope.ServiceProvider.GetRequiredService<ReconciliationAgent>(); 
       var result = await agent.RunAsync(); 
       _logger.LogInformation(result); 
       await Task.Delay(TimeSpan.FromMinutes(2), stoppingToken); 
     } 
   } 
}

Register it:

builder.Services.AddHostedService<AutonomousAgentService>();

Visual Studio Structure

Swagger / API Running

SQL Before Agent Execution

Background Logs

SQL After Agent Execution

What Makes This “Agentic AI”?

Traditional Automation	Agentic AI

Hardcoded workflow	Goal-driven reasoning

Manual trigger	Self-running

Static rules	Context-aware

No explanation	Natural-language audit trail

Summary

In this POC we moved from:

API → Automation → Intelligence → Autonomy

We built a system that:

✔ Understands a goal
✔ Uses tools dynamically
✔ Fixes business data
✔ Runs continuously
✔ Explains its decisions

This is the architectural shift happening across modern enterprise platforms.

Agentic AI is not replacing applications — it is becoming the decision layer inside them.