August 9, 2026
Bot wrote it? Prove it
Human vs. AI – Diff-based line-level provenance for text under agentic editing
A new tool wants to expose which lines are yours and which were written by the bot
TLDR: A new tool tries to label which parts of a document were written by a human and which came from AI, so future bots know what not to bulldoze. Commenters loved the idea but immediately argued over messy edge cases, especially whether it works at all if the AI’s edits are saved under your own name.
The latest mini-drama in AI coding land is wonderfully petty: who actually wrote this thing? The project, Us vs. Them, promises to scan a file’s history and mark which parts were written by a person and which came from an AI helper. In plain English, it’s trying to stop your chatbot from barging back in later and rewriting the precious bits you personally sweated over. Think of it as digital turf protection for your notes, app, or README.
And yes, the comments instantly turned this into a custody battle over code. One camp was basically: nice idea, but we already use labels for this. Gritzko chimed in that annotations work fine—until the AI ignores them “1 out of 100 cases,” which is exactly the kind of tiny failure rate that becomes nightmare fuel when you’ve got enough files. Another commenter raised the bar with a hilariously strict definition of human ownership: if AI wrote it but a person changed even one character, congrats, now it’s human-touched. That’s the kind of loophole logic the internet lives for.
But the spiciest pushback came from people asking whether the whole thing falls apart in real life. If your AI assistant is working on your own machine, one commenter noted, aren’t all the saved versions still under your name anyway? Translation: if the bot wears your name tag, can this tool really tell “us” from “them”? That doubt gave the thread its main tension: brilliant safeguard, or neat idea smashed by messy reality?
Key Points
- •The article introduces "Us vs. Them," a diff-based system for line-level provenance in text edited by humans and AI agents.
- •The approach is designed to work on plain text without special markup by relying on version history with identifiable authorship.
- •It outputs ranges of text with authorship scores, describing human-authored sections as islands within machine-generated content.
- •The algorithm is based on simple diffing but is intended to preserve coherent text regions and account for authorship dilution and structural changes.
- •The tool can be used as a library or CLI, including inside git repositories, with parameters to define which authors are treated as human or agent.