Building THE WHEEL
What it's actually like to build a private knowledge engine from scratch.
On building an AI assistant that knows which capability to use, when to ask for more context, and when the smartest thing it can do is stop.
Privacy was never about hiding. It was always about choosing. We built an entire regulatory infrastructure around protection and never built the architecture for the other half: the right to expose on your own terms.
51 messages. 6 hours. 3 PRs. And one human who kept saying “that doesn’t make sense.” What AI debugging collaboration actually looks like, from the transcript.
A debugging story about a bug that looked like a cross-user privacy disaster, why it had to be treated that way, and how careful tracing proved the core system was actually working.
What building a design system before launch taught me about consistency, speed, and the strange intimacy of writing every word your users will ever read.
Nobody had solved multi-tenant semantic search over encrypted data. Here’s the journey across four fields — and a bit of Doctor Who — that found an answer.
What 300+ stories, 8 workflow versions, and 36 shelved feature branches taught me about building something new. Under 10% of everything I built will ship at launch. That's the point.
AI doesn't raise the ceiling on what experts can do. It raises the floor on what everyone else can explore. What building solo with AI actually looks like — and what it means for innovation.