syncd-brain, the company brain
A company of two collects more research, open questions and half-made decisions than two heads can hold, so all of it lives in one repository where AI agents do the legwork.
Most of the work went into the harness around those agents, because an agent is a model plus its harness. It has custom skills for each field we work in, and a knowledge base where the claims we use in public are checked against their primary sources.
Amber and I get open questions one at a time, each with the context to answer it, and nothing decided gets lost.
- harness engineering (HEAD -> main)agent = model + harness
- review loopwhat fails goes back
- agenticagents do the legwork
- context engineeringthe right files at the right time
- prompt engineeringwhere it started