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 April 10 - April 16, 2026

What I Worked On

Autonomous mode can run your backlog. But a backlog is not a plan. This week I closed that gap.

The biggest push was turning the issue tracker into a real planning surface. I added Milestones, hard due dates, and effort, so work has shape and a deadline instead of being a flat list. I built a My Week view that shows what is actually on your plate right now, filtering out issues that are actively claimed by someone else so you see your work, not everyone's. Dashboards got faster, with lazy-loaded closed issues and accurate child counts, because a planning tool you dread opening is a planning tool nobody uses.

Then I made autonomous mode match that maturity. It used to only run against Epics. Now it runs any issue type that has child issues, it supports issues that get skipped mid-run, and there are per-worker kill and restart controls, so if one worker goes sideways you stop that one without tearing down the whole run. That is the difference between an autonomous system you trust and one you are afraid to start.

What the AI Did Well

The agent built the Milestones and due-date system and the My Week view as clean, wide features, and it took on a big documentation job: overhauling the framework docs down to a focused API reference with XML remarks on every core interface. It also did the security passes I never want to skip: file-upload validation, an AnyAdminAccess policy on the admin API controllers, and input sanitization on the waitlist admin. I even used the new skills to have DevNitro generate a marketing content calendar for me, which is the platform doing my own non-code work.

What Needed Correction

A model update caused a scramble. When Claude updated, it started wanting to call tools it had not before, so I built a monitor tool and a schedule_wakeup tool to give it sanctioned ways to do what it was reaching for, instead of fighting it. I also hardened the tool-call path: malformed tool-call JSON now gets a parse-error message back so the model can correct itself, and provider-side response truncation gets detected instead of silently corrupting a turn.

I found a bug where due dates and effort were not saving correctly, right as I was trying to use the new planning features for real. Nothing surfaces a bug like trying to depend on the thing you just built. Fixed.

What Shipped

DevNitro has Milestones, due dates, effort, and a My Week view, with faster dashboards. Autonomous mode runs any issue type with skip support and per-worker kill and restart. The GCT framework docs are a clean API reference. And the tool-call path handles malformed JSON and provider truncation, backed by a monitor tool and a scheduled-wakeup tool.

The Numbers

Category	Hours
Product / Features	30h 34m
Content	7h 58m
Docs / Skills	6h 9m
Business	3h 12m
Community	2h 15m
Total	50h 7m

Six hours in docs is unusually high for me, but it was worth it. A framework nobody can read is a framework nobody adopts, and the same is true of a backlog nobody can plan around. But it is even more important when agents need to understand how the framework functions so they can use it correctly. The docs are not just for humans, they are for the AI too.

The promise of Intelligent Engineering is not just an agent that writes code. It is a system your team can trust, support, and scale. Trust starts with a plan you can see and an autonomous run you can steer. This week was about both.

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

April 17, 2026 Build Log: Turning the Backlog Into a Plan