Late last year I started building amoOS as a personal knowledge base: a place to forward links, capture notes, and search everything I'd ever learned. That part still exists. But somewhere along the way the project inverted, and the inversion is the interesting part.
The realization was this: once AI agents can write most of the code, the scarce resource stops being engineering hours. It becomes deciding what to build, verifying that what got built actually works, and remembering everything the process learned. So that's what I optimized. Today amoOS runs every project I own, eleven of them at last count. A fleet of AI coding agents does the building. I plan the work, dispatch it, and verify the results. The system keeps the memory: every session, decision, and forwarded link gets distilled into a living knowledge base that currently holds over four hundred compiled pages.
It reports to me through three surfaces: a native macOS cockpit for deep work, Telegram when I'm away from the desk, and a web view for reference. And it measures itself: utilization, yield, attention cost, and how fast it can produce a stakeholder-ready brief on any project. If the factory can't account for what it shipped this week, that's a bug.
The rule that makes it all work is boring: nothing ships without human verification. Autonomous output without trust is worthless. 'Verified done, with evidence' is the only unit of progress that counts.
Why does a solo builder need all this? Leverage. Salesbugle, client MVPs, this website, and amoOS itself all run through the same machine, and every lesson learned in one product immediately benefits the next. I'll be sharing what I learn as I push this further, the wins and the failures.