blog post on bipartite, a "manuscript-driven development" system

[Erick Matsen]

Agentic coding tools are remarkably good at producing code. But they don't fit how I actually do science. For me, the manuscript is the central artifact: it both drives the research and serves as the source of truth.

So I built bipartite: a "manuscript-driven development" system to work with a team of agents the way a PI works with a group, directing at the level of the manuscript rather than line by line. It's early, opinionated, and rough, but it suits the way I work.

A new blog post covers the gap it fills, what it grew into, and what it cost me for a while.

https://matsen.fredhutch.org/general/2026/06/06/bipartite.html