Field notes.
Live agents.
Real receipts.
The working notebook of BJGLLC — how a one-person practice runs on AI agents. Overnight fleet runs, orchestration patterns, the systems underneath, and what actually shipped. First person throughout; AI involvement disclosed on every post.
The wall.
16 tasks, 47 subagents, zero humans: the overnight build
A 16-task plan, 47 subagents, and two models from two vendors adversarially reviewing each other overnight — what the structure caught while I slept.
16 tasks, 47 subagents, zero humans: the overnight build
Private-first, public-shaped: designing our first public product
Two days between a manual fleet sweep and a buildable spec: the market check, the four design decisions that mattered, and the dogfood gate behind muster.
Private-first, public-shaped: designing our first public product
The task manager didn't fail. The task shape changed.
Agents invented a category of task that Todoist and Linear had no place for. The story of the gap, the manual sweep that named it, and the layer that grew underneath the stack.
The task manager didn't fail. The task shape changed.
What this is, and isn't.
This is a working notebook, not a content farm. Every post documents a real run of the BJGLLC agent stack — the sessions happened, the transcripts are archived, and the numbers are receipts, not estimates.
Posts are written in first person and every post states exactly how AI was involved in the work and in the writing. The site itself was built by the same agent fleet it writes about.
Built on the BJGLLC design system · Atlanta, GA · 2026.