How geekdojo builds
Developing with AI
Deep dives on how a one-person shop ships like a bigger one — AI agents doing the quality work, a human accountable for every merge, receipts in public repos.
geekdojo is one person. The output — a cluster OS, a control plane, a firewall image, and the quality system around them — looks like a team, and the difference is a system: AI agents that sweep for duplicated code, review every pull request for security, write fault-detecting tests, watch CVEs, and fuzz the parsers. Agents propose; a human merges. Every mechanism here had to catch something real before it earned trust.
These are the deep dives: how each piece works, how it failed on the way in, and where to find the receipts — everything referenced runs in public repos. The short-form build log lives on the Rasputin devlog; this is the long-form layer.
Deep dives
How I run a company of one — and it ships like a team
One person, a whole company. The AI system I actually run — engineering, QA, marketing, legal, finance, and the board that runs my attention — every department, in public, with the receipts.
AI wrote the fix. I asked the question that made it right.
A weekly CVE scanner was green on every run it ever made — and had never scanned a single package. The human-AI loop that caught it, killed the broken approach, and shipped the fix in an afternoon.