Principles

The lab runs on a short list of rules. They are written down here so they can be checked against later.

​

One researcher, many agents

Everything shipped under the lab is built by one person working with AI agents — research, drafting, engineering, deployment. The constraint is deliberate: it forces products to stay small enough to fully understand. Anything that would need a team to babysit does not get built.

​

Boring infrastructure, open protocols

Static sites. Cloudflare free tier. RSS, sitemaps, JSON-LD, MCP. Nothing that requires an ops rotation, and nothing that locks content into a proprietary silo. The payoff is cost near zero and failure modes that are boring too.

​

Small bets, real data

A new project starts on a free subdomain and has to show a real signal — paying users, steady traffic, a customer naming it — before it earns a domain. Tools are built only where the data to feed them actually exists. No product ships with demo numbers standing in for real ones.

​

Maintain or retire, explicitly

Projects are either maintained or retired, and retirement is announced rather than left to rot. The project index carries that ledger. Archived projects stay online where they still serve readers, but nothing pretends to be alive.

​

Docs in the open

Every product documents itself on its own domain, and the lab exposes its content through read-only MCP endpoints so agents can read the same thing humans do. If a claim cannot be linked, it does not go in the docs.