Building WinUI Apps with Claude Code: AI-Driven Windows Desktop Development with the WinUI Agent Plugin and GitHub (C#/.NET in Practice Book 3)

Ask an AI to build me a WinUI 3 app and, somehow, the code it hands back won't run — this book crosses that wall with Microsoft's official WinUI Agent Plugin, and takes you from an idea to a distributable Windows desktop app, AI-driven from start to finish.

A general-purpose AI is pulled toward older WPF and UWP styles and tends to emit code that won't run in WinUI 3. The WinUI Agent Plugin is Microsoft's official plugin for giving an AI the correct WinUI 3 shape. This book uses it inside Claude Code (GitHub Copilot CLI and OpenAI Codex also work).

The subject is a single Pomodoro timer. From an empty repository, clearing GitHub Issues one at a time, you grow the same app all the way through testing, signing, and distribution.

What you'll learn - The standard AI-driven workflow (Issue → plan with /kickoff → hand off implementation → /wrapup for a PR → confirm yourself → merge)

- Setting up your environment (winapp CLI, WinUI 3 templates, Claude Code, Microsoft Learn MCP Server)

- Building the app (a two-screen timer as a single Issue; you write the what )

- Quality (anti-pattern checks with winui-code-review, automated UI tests with winui-ui-testing)

- Shipping it (MSIX packaging, code signing, Microsoft Store submission)

What matters most here is don't just hand everything to the AI. You keep control of the development and hand only the tedious implementation to the AI — that division of labor is the book's backbone.

Who this is for - Developers with the C#/.NET basics who want to build a Windows app together with an AI

- Anyone who wants to experience AI coding (Claude Code and the like) end to end on a concrete subject

- Anyone who tried to build an app with AI but couldn't get it to run the way they wanted

C#/.NET in Practice series This book is part of the C#/.NET in Practice series. Its companion Building Desktop Apps with C# and WinUI 3 gives you the foundation to judge the code an AI generates. Each volume stands on its own.

The template and finished sample are on GitHub.