us-vs-them: Line-Level Human vs. AI Authorship
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- us-vs-them is a CLI tool and library that diffs a git repository's version history to score line-level human vs. AI authorship, with outputs ranging from 0.00 (fully agent) to 1.00 (fully human), including fractional scores like 0.46 for partially modified text
- The tool targets agentic editing workflows where developers want to protect their own writing — e.g., opening paragraphs of a README.md or key code sections in a mostly vibecoded app — from being overwritten by agents in future sessions
- us-vs-them works on plain markdown/text files with no markup required, relying solely on the fact that each new version is committed under identifiable authorship
- The output takes the form of "islands" of human-authored lines inside a "sea" of machine-generated text, with the algorithm factored to handle joining, splitting, and dilution of authorship without collapsing to fully one side or the other
- Configuration uses
--oursto name humans or--theirsto name agents (whichever list is shorter), with both flags at once rejected; installation usesmake installvia bbin
Why it matters: As agentic coding sessions become routine, developers have no built-in way to tell an agent which lines of a file are their own. us-vs-them offers a markup-free, git-native approach by treating commit history as a provenance trail, producing fractional authorship scores (0.00 to 1.00) that let a human flag protected zones — a concrete mechanism for asserting ownership over islands of human work in an AI-generated sea.
Ask SkimNews
