A note about this build log entry: I published this after the week it covers. DevNitro reconstructed a draft from work-related entries in my journal, Git commit history, and time tracking. I reviewed, edited, and approved the final post. The story reflects what I knew and built at the time.

From the Backlog to the Build Log

Week of January 30 - February 5, 2026

What I Worked On

This was a refactor week, and not a small one. 114 commits, almost all of them structural. Before I put this in front of paying customers, I wanted the codebase speaking one language, so I spent the week driving three .NET architecture patterns through DevNitro, GCT, and BlankSite all at once.

First, IOperationResult. Every service and API now returns a typed result that says success or failure with a reason, instead of throwing exceptions or passing loose message strings around. Second, a UiService pattern that aggregates the data a dashboard or form needs behind one call, with a single unit of work per request. Third, dual-mode services, so the code works seamlessly in both server-side rendering (SSR) and WebAssembly (WASM) — the same Blazor component code runs identically regardless of render mode. The service runs on the server, but also has a client implementation that calls the controller via HTTP. The controller is simply a thin wrapper around the service, so the service is the single source of truth for all business logic.

The big data change was migrating every entity ID from int to long. That is a 64-bit future-proofing move that sounds trivial and touches almost every file, table, and test in the system.

I also built out workstreams, the capacity model that tracks concurrent AI work, metrics, and usage in billing orgs.

What the AI Did Well

This is the kind of wide, pattern-driven change AI is built for. Migrating hundreds of call sites to operation results, threading long IDs through entities and tests, and converting services to dual-mode is repetitive and unforgiving, and the agent held the pattern across all three repos. It's times like this I wonder how any serious software can be written and seriously improved in a non typed language. (Apologies for my bias here.) I ran multiple instances of DevNitro simultaneously, working multiple codebases at once to apply the refactors.

I added a few real tools this week too: a WebSearchTool so the agent can search the web, and a Conversation Viewer so I can browse chat history and see tool calls grouped sensibly.

What Needed Correction

Honesty first: the refactor was much bigger than I estimated. I planned to film a demo and could not, because the change kept sprawling. Turtles all the way down.

The autonomous runs kept stopping at 150 iterations, so I raised the loop limit. And I blew through my monthly model plan partway through the week and jumped to the next tier to keep going. Heavy refactoring using AI is not free. Still, it is worth it for the long-term maintainability and consistency it brings to the codebase.

What Shipped

DevNitro, GCT, and BlankSite share these same three patterns now: operation results, aggregated UI services, and dual-mode Blazor services. Every ID is a 64-bit long. Conversations persist on the server, with local-only sessions supported. Workstream capacity tracking and billing are in, along with a Products-driven pricing model and annual billing. And the agent can search the web and browse its own conversation history.

The Numbers

Category Hours
Product / Features 58h 39m
Business 3h 55m
Content 3h 32m
Community 1h 7m
Total 67h 14m

Sixty-seven hours, and almost all of it structural: 24 on code refactor, 10 on the database migration, plus sessions, workstreams, and issues. A foundation week, not a feature week. (I just saw kids saying 'six seven six seven' in my head)


Nobody buys software because of its .NET architecture patterns. But consistency is what lets me hand a codebase to an agent and trust the result. Plus, when the blueprint is going to be the basis of many future projects, it pays to get it right early.

Curious how this site was built?
Watch the 5-minute walkthrough at devnitro.com →

An AI built this blog (but not this post)

This entire blog system (domain models, database, services, UI, auth, etc.) was built by DevNitro on its own in 95 minutes. I recorded every second.

    We respect your privacy. Unsubscribe at any time.