Birthday release day: parallel agents, a remodeled MVP, and calmer judgment
Used my birthday as a delivery checkpoint: coordinated parallel agents, moved a client MVP through backend, tenant flows, catalogs, invitations, planning, and visual review.
My birthday became a release day. Not in the noisy sense. More like a checkpoint: enough parallel work, enough review, and enough product judgment to see the system taking shape.
Client project (NDA)
Most of the day went into the remodeled MVP. I used several Conductor workspaces in parallel and kept collapsing the work back into one product story. The backend schema was rebuilt around the target model, with migrations and seed data that match the new direction. Authentication, sessions, company selection, backoffice creation, invitations, tenant navigation, and role-aware access all moved forward.
The front end started to look less like isolated screens and more like an application shell. I spent time testing the real UI, checking which roles should see which areas, and making sure management flows did not leak into pilot workflows. That kind of permission detail is boring until it is wrong. Then it becomes the whole product.
Product operating system
A lot of the day was not code. It was deciding how to run the work. I reviewed which issues were actually done, fixed missing closing links on pull requests, cleaned labels, planned what could run in parallel, and separated tasks that a junior developer could take from tasks I should keep.
I also revisited the relationship between product requirements, specs, issues, and architecture decision records. The lesson was simple: issues are not the source of truth when the spec carries the real context. The system works better when the issue points to the spec, the spec carries the decision, and the pull request closes the loop.
Tooling and review
I documented how to use Agent Browser in the project and used visual testing as part of the review cycle. The goal was not to collect tools. It was to make agents more useful inside a real product workflow: test the UI, compare the screen to the intended shell, review the pull request, merge back into main, and keep the next agent from starting on stale ground.
Broddy
Broddy stayed in research mode. I looked at React Native AI, edge inference, and small model work to understand what belongs on-device and what should stay outside the app. That question matters because the assistant only feels personal if it is fast, private by default, and cheap to run.
The best birthday gift was not a finished product. It was seeing the process mature: parallel agents, clearer specs, tighter reviews, and a little more calm in how I choose the next move.