The Legacy .NET Problem
Across enterprise software, there are thousands of thick-client .NET applications — WinForms, WPF, classic ASP.NET — that contain 10–20 years of business logic. These systems work. The companies that built them are not going to throw them away.
But the market is demanding AI-native capabilities: intelligent document processing, autonomous workflows, conversational interfaces. Legacy stacks can't deliver these without significant engineering investment.
The conventional answer — rewrite everything — fails 70% of the time. Projects go over budget, over time, and often produce worse systems than what they replaced.
The Better Approach: Semantic Kernel
Microsoft's Semantic Kernel is a .NET-native SDK for building AI-powered applications. The key insight: you can inject Semantic Kernel into an existing .NET application without rewriting it.
var kernel = Kernel.CreateBuilder()
.AddAzureOpenAIChatCompletion("gpt-4", azureEndpoint, apiKey)
.Build();
// Your existing business logic stays. AI gets added on top.
var result = await kernel.InvokePromptAsync(
"Summarise this patient record: {{$record}}",
new KernelArguments { ["record"] = existingPatientData }
);The Migration Framework
Phase 1: API Layer (Weeks 1-4)
Wrap core business logic in REST APIs. Don't change the logic — just expose it. This enables new frontends (React, mobile) without touching the engine.
Phase 2: AI Augmentation (Weeks 5-12)
Identify highest-ROI AI injection points: document processing, search, summarisation, data extraction. Implement with Semantic Kernel targeting these specific workflows.
Phase 3: Agent Workflows (Months 4-6)
Replace manual approval chains with supervised agent workflows. Humans review agent decisions initially, then step back as trust builds.
Result: Thick Agents
The legacy system still runs the business. The AI layer handles intelligence. The result is a "thick agent" — a system with the deep business logic of 15 years of enterprise development, augmented with autonomous AI capabilities.
This is the only viable path for most enterprises. Full rewrites are a fantasy. Augmentation is achievable.