Last post I talked about what I’m building and why. This one’s about the how.

I’m building a real estate property management website from scratch. Full stack - backend API, database, frontend, hosting, the whole thing soup to nuts. Here’s what I picked and why.

.NET 10 for the backend. I’ve been in the Microsoft ecosystem my entire career so it’s not only what I know and it’s what I’m fastest in. Also, it wouldn’t be Agentic .NET very much without it. I think this will provide a nice baseline for speed as the better you know your stack, the better the prompts and output should be. I’ll be able to move the fastest and catch mistakes the fastest.

Cosmos DB for persistence. This is a little stretch for me. Most of my career I’ve been in SQL, so I want to move away a bit for this agentic project. Having Cosmos DB as the document database — NoSQL with Azure-native fits as the data model here. A property has tenants, documents, and requests for maintenance. That nests cleanly into a document database instead of spreading across relational tables. I’ve used Cosmos once, so it will help further gauge the value of agentic driven development.

React, TypeScript, and Tailwind for the frontend. Now a bigger stretch because I rarely have done any frontend work professionally. React and TypeScript have massive ecosystems, and the agentic tooling for both is excellent. I’ll be able to really judge the power of agentic spec-driven development with this approach. I threw in Tailwind CSS as a utility-first styling framework. Because of composing styles directly on the elements it will be fast and I expect this to help when AI is writing the UI.

Azure App Services for hosting. I’m already deep in the Azure ecosystem at work. App Services gives me a managed platform for the API without the overhead of managing infrastructure. For the kind of app I’m building — authentication, persistent database connections, middleware — App Services will be the right fit.

Fully Agentic. Every single piece of this project will be built with an AI agent as a collaborator. Not a copilot. Architectural decisions, implementation, testing - everything should be agentic. Let’s see what agentic development can do in .NET with a greenfield project.

Next time I’ll get into what the setup looks like as I kick off the project. I’m interested in seeing how the agentic loop goes, what turns out well and what doesn’t when building something real.

Keep reading