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 October 31 - November 6, 2025

What I Worked On

Shorter week. We drove out to a college expo up at the Ark Encounter and Creation Museum at the end of the week, so the build time was front-loaded into the first few days. The main thread was dogfooding: I used DevNitro to generate my own site, HowdyChad, from the BlankSite blueprint. This is AI-assisted .NET development pointed at my own backlog instead of a demo.

That is the test I care about. If DevNitro can stand up a real site I plan to run in production, the platform is doing its job. I had its MCP tools create HowdyChad, wired it to a SQLite database, and ran the migrations against it. SQLite is deliberate. For a content site like HowdyChad I want the cheaper, simpler database, and I am the first customer asking for that option.

Along the way I found BlankSite's data stores were not using the latest unit-of-work Table and Column pattern. I fixed that in BlankSite and in DevNitro so generated sites inherit the current pattern instead of an old one.

What the AI Did Well

Generating HowdyChad through the MCP tools mostly worked. DevNitro created the site, and getting SQLite connected and migrated went smoothly enough that the friction moved to the parts that were not built yet, like outbound email (user registration flow).

I also had it clean up years of formatting drift in GCT. Normalizing line endings and indentation to the .editorconfig and .gitattributes rules is tedious, mechanical work, and a good fit for an agent to handle via scripts. I folded in a round of code-review fixes too: a bad UPPER() call on the wrong side of a column comparison, corrected T4 template database names, and schema and ORM updates.

What Needed Correction

The seed-data isolation bug was the interesting one. My prototype's tests kept expecting to see GCT's own test data. The AI made the same mistake because it could see the GCT source alongside the site's code, so it assumed the two shared a seed. I reworked the seed data so each site owns its own, which is correct no matter what the AI can see. Giving an agent more context can also give it more ways to be wrong.

Email is not done. HowdyChad needs to send mail through an Azure queue that is not set up for generated sites yet, so that carries into next week.

What Shipped

HowdyChad exists (on disk) as a generated site with a working SQLite database and migrations. BlankSite added a waitlist feature. The unit-of-work Table and Column pattern is consistent across BlankSite and DevNitro. GCT is normalized to its formatting rules with a round of review fixes in. On the business side, I pointed my old marketing domains at a maintenance page in Azure and set up redirects toward devnitro.com, so the brands finally point in one direction.

The Numbers

Category	Hours
Product / Features	19h 11m
Content	9h 37m
Community	1h 3m
Business	0h 30m
Total	30h 22m

The lighter total reflects the travel at the end of the week. Most of the product time went into generating HowdyChad and fixing the store pattern, with content hours on the weekly videos and a small slice on go-to-market planning.

Using DevNitro to build the site I write on is the real test of AI-assisted .NET development. Every rough edge I hit here is one a future customer will not have to.

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

November 7, 2025 Build Log: Building HowdyChad With DevNitro