Founder OS

On a growing team, the founder can't be in every thread. People either wait — slow — or guess, and the company drifts. Most 'AI persona' tools make it worse: they answer everything with the same confidence, so the first time they're wrong, trust is gone.

An async stand-in for a startup's own founder. It answers in the founder's voice when it has grounds to, escalates a crisp one-line decision to the real founder when it doesn't, and learns from every correction — compounding a decision memory that keeps the company's judgment available even when the founder isn't in the room. Low confidence isn't a failure; it's the product's second job.

Vibe-coded at a Vercel hackathon in San Francisco: it started as a celebrity-archetype toy, and I pivoted it mid-event into the real product — a working, deployed version in about 40 minutes, demoed to the room. Next.js on Vercel, the Vercel AI Gateway, Neon Postgres, and Claude generating the structured answer card (Take · Principle · Pushback · Action · Confidence). I later rebuilt it with Claude Code on a current stack — AI SDK streaming, Drizzle, Zod, pgvector — with a warm, considered interface, because the emotional job is to make asking a 'dumb' question feel safe.

An idea taken from zero to a live, testable product with AI tooling in hours — and the product instincts that make a prototype matter: escalation-as-a-feature, a compounding decision memory, and an interface designed so hard questions are easy to ask. The public playground answers as founders like Brian Chesky, Sam Altman, Dario Amodei, and Jensen Huang.