From the Backlog to the Build Log

Week of July 17 - July 23, 2026

What I Worked On

This was a positioning week with a sharp product edge. Most of my hours went to content. I published the "What Is Intelligent Engineering" post and its video, and wired every build log since April back to a new /intelligent-engineering page. The idea is that the AI a team runs needs an operating model around it, not just faster code generation, and now it has a home page instead of living only in my head.

The product work was light this week, but it still mattered. I built a conversation recap feature into the DevNitro API and CLI so a new session can pull a recap of an earlier conversation instead of starting cold. I built it for a selfish reason: I use it to draft these build logs. I also taught the ConversationCommand to return a single day of conversation and fixed the dashboard so it switches context when you open an issue in another project.

The rest of the week ran through the apps I build with DevNitro. The HowdyChad blog got a publish endpoint and promote support in its MarkdownPublisher, which is how I post these build logs from the command line. Its editor also learned to keep HTML tables intact, so a Numbers table like the one below survives an edit in the Quill editor. I refocused the GlobalCove site on Intelligent Engineering positioning, adding a new hero, a two-option contact flow, and analytics.

What the AI Did Well

The recap feature came together as a clean vertical slice across the API and CLI, and DevNitro fixed its own bug once I found the cause (more on that below). The NuGet upgrade sweep across GCT, DevNitro, and BlankSite went smoothly, with one test expectation to nudge. The tool-call parser also learned to read flattened JSON, so a model that emits its call in a different shape no longer trips the run. Small, scoped changes with tests are where AI-assisted .NET development earns its keep.

What Needed Correction

Two things bit me, and both are worth telling.

First, the recap tool failed the first time I used it for real. I pulled the server logs, handed them to DevNitro, and the root cause was a missing database index. Adding the index fixed it, and I made the HttpClient timeouts configurable while I was in there.

Second, the payment work was a regression, not a new feature. A security fix from last month's sweep broke new-card checkout in GCT for a particular processor. Existing cards kept working, so the break hid for a couple of weeks until a new customer had a checkout fail. The fix needs an end-to-end test that runs a new card through checkout, which is now in the backlog for the next sprint.

What Shipped

The "What Is Intelligent Engineering" post and video are live, and every build log since April links to the Intelligent Engineering page. Conversation recap is live in the API and CLI, with single-day retrieval in the ConversationCommand and cross-project context switching in the dashboard. GCT card payment regression from a security fix was addressed. Package versions across GCT, DevNitro, and BlankSite are current. The HowdyChad blog gained a MarkdownPublisher publish endpoint (I post build logs by CLI now) and preserves HTML tables in its editor. The GlobalCove site now leads with Intelligent Engineering positioning, with a new hero and a two-option contact modal, and it offers an Intelligent Engineering Sprint and Intelligent Engineering Workshop. The MarkdownPublisher (which can also handle uploading images to Azure Blob storage) was securely run without the agent ever having the API key because of the Lobast integration work from the week before.

The Numbers

Category	Hours
Content	29h 57m
Business	8h 38m
Product / Features	1h 54m
Community	1h 11m
Deploy / Infra	50m
Total	42h 30m

Content and business took about 38 of the 42 hours. That is a positioning week, and I own it. The product hours were few, but the recap tool and the checkout fix are the parts a customer would feel.

Where This Fits in Intelligent Engineering

The recap failure provides a lesson for us. The code looked fine and passed as a feature, then fell over on first real use because of a missing index. Intelligent Engineering checks the outcome, not the diff. You prove the thing works under production conditions, because a team cannot support what only ran on a developer's machine.

Most of this week went to explaining the work. The rest went to proving my own tools run. I'm refraining from continuing to build new features until I have more user feedback.

Curious how this site was built?
Watch the 5-minute walkthrough on YouTube →

July 24, 2026 Build Log: Conversation Recap and Payment Security