From the Backlog to the Build Log
AI-assisted .NET Development, August 7 to August 13
What I Worked On
This week, AI-assisted .NET development meant fixing the parts around the code, not just writing more code. I spent most of my DevNitro time on tool reliability, visual browser cleanup, tests, security, and the path from a blueprint to a checked build.
That is the work behind Intelligent Engineering. An AI agent can produce a lot of code. The system around it must make sure the work ends, cleans up after itself, and proves the result.
The largest technical thread was the visual agent browser lifecycle. A browser could remain open after a run or a crashed session. That wastes memory and makes long-running sessions less stable. I worked through a set of backlog items for process tracking, exit hooks, idle cleanup, forced shutdown, startup cleanup, and end-to-end proof that the leak was closed.
I also tightened tool behavior. DevNitro now avoids duplicate read-only tool results, resets tool state after context loss, handles bad retry IDs more clearly, and retries short Git configuration failures. Blueprint checks gained support for subfolder roots, clearer verification, and separate compile and run timing.
What the AI Did Well
The browser cleanup work was a good fit for backlog-driven agent work. The tasks had clear bounds and a result that tests could prove. The agent implemented the pieces across process tracking, browser disposal, app shutdown, idle sessions, and crash recovery.
It also handled a mix of smaller work without losing the main thread. That included safer sign-in failure messages in GlobalCove, reserved user status rules, wall-clock limits during merge checks, and a billing address value object. Across DevNitro, GCT, BlankSite, GlobalCove.com, and HowdyChad.com, the week produced 42 commits.
The product and its own sites moved together. I upgraded the blueprint package, added blueprint pages, changed the DevNitro site to match the current product story, and added consent-gated analytics to my own sites. The same blueprint and rules used for customer work had to hold up on my work.
What Needed Correction
The week included an intermittent test failure around very short file time checks. The first approach depended too much on clock resolution. I replaced that timing work with Stopwatch and made the comparisons more precise where needed. The test caught a real weak point before it became a harder production bug.
I also had to keep tightening failure paths. Browser disposal needed more than a normal close call. It needed a time limit and a forced shutdown path. Tool state needed an explicit reset when a session lost context. These are not flashy changes, but they decide whether an autonomous run stays reliable after days of use.
What Shipped
The visual agent now tracks browser processes, cleans up idle browsers, closes browsers when apps stop, and sweeps up processes left by crashed sessions. End-to-end checks prove the browser leak is closed.
DevNitro also shipped stronger tool deduplication, state reset, retry handling, event storage, blueprint verification, and timing data. GCT gained safer sign-in rules and error messages. The DevNitro and HowdyChad sites moved to the latest blueprint preview, while the GlobalCove site gained consent-gated tracking and a lead capture flow.
The Numbers
| Category | Hours |
|---|---|
| Product / Features | 10h 6m |
| Content | 8h 18m |
| Business | 7h 21m |
| Deploy / Infra | 2h 34m |
| Testing | 2h 17m |
| Total | 30h 37m |
Product work led the week. The business time covered go-to-market work. Content covered the HowdyChad blog, stories, and YouTube work. Client work and personal entries are not part of these numbers.
Where This Fits in Intelligent Engineering
Visual verification cannot become a trusted part of AI-assisted .NET development if every run can leave a browser behind. The feature must include cleanup, failure handling, and proof. That is how a team gets software it can trust, support, and scale.
The code was only part of this week's work. The rest made sure the code could run again tomorrow.
Curious how this site was built?
Watch the 5-minute walkthrough on YouTube →
0 Comments
No comments yet. Be the first to share your thoughts!
Leave a comment