# How We Use Proof, a Collaborative Editor for Humans and AI Source: https://www.youtube.com/watch?v=P9XfYZh9CwA Recap page: https://rapidrecap.app/video/P9XfYZh9CwA Generated: 2026-03-11T15:32:46.628+00:00 --- ## Quick Overview The team uses Proof, a live collaborative editor for humans and AI, to streamline their workflow, allowing agents to read, comment on, and update documents in real-time, which significantly improves efficiency and clarity over traditional methods like Google Docs or separate code/Notion tasks. **Key Points:** - The team requires a live collaborative web-based document editor for humans and AI agents to work together as equals. - Proof enables agents to suggest edits, leave comments, and track every character's provenance (who wrote what). - The tool works in the browser without needing installation, allowing users to share a link and start writing immediately. - The system is designed to prevent human agency erosion by keeping humans in control while leveraging AI for suggestions and edits. - The development process involved iterating on an initial Mac app version, eventually moving to an open-source, web-based editor. - The process is highly collaborative, with agents providing real-time feedback on documents, which is then reviewed by team members like Dan and Rachel. - The core philosophy is to make writing production faster, clearer, and more engaging by integrating AI seamlessly into the document flow. ![Screenshot at 0:06: The screen displays the "Proof Launch – Podcast Agenda" document in the Proof editor, clearly showing the collaborative nature where different participants \(represented by avatars on the right\) are working on the same document, outlining the problems Proof aims to solve.](https://ss.rapidrecap.app/screens/P9XfYZh9CwA/00-00-06.jpg) **Context:** Dan Shiper, CEO of Every, introduces Proof, a collaborative document editor designed specifically for human-AI collaboration. The discussion centers on the necessity of such a tool to solve the problems of traditional AI writing workflows, where transparency and control are often lost. The video features Dan explaining the tool's core features and philosophy, with input from team members Brandon Gell (COO of Every), Kieran Klaassen (GM of Cora), and Austin Tedesco (Head of Growth at Every). ## Detailed Analysis The discussion revolves around Proof, a collaborative editor built for seamless interaction between humans and AI agents. Dan Shiper explains that the fundamental problem with current AI writing tools is the lack of collaboration and transparency, often resulting in humans either controlling the AI too much or losing agency entirely. Proof solves this by providing a live, web-based document editor where humans and AI agents work as equals. Key features include: AI agents suggesting edits and leaving comments (which humans can accept, reject, or reply to), provable provenance tracking for every character written (human or AI), and zero login required for reviewers to access the document via a shareable link. The team found that integrating AI directly into the document flow, rather than using separate chatbots, was crucial. They initially developed a Mac app but transitioned to a web-based, open-source product. A key takeaway from their internal process is that having agents assist in structured tasks like outlining and summarizing (as demonstrated by the R2-C2 agent in Slack) is highly valuable, especially when the human collaborator (like Dan) can then iterate on the AI-generated structure. The tool aims to accelerate production speed and allow for a more intuitive, continuous feedback loop, contrasting sharply with the clunky nature of using tools like Google Docs for this purpose. They aim for a future where the AI acts as a helpful co-author rather than an opaque black box. ### The Problem - Writing with AI is broken due to lack of collaboration and transparency, leading to either human overcontrol or loss of agency, unlike human-human collaboration solved by tools like Google Docs. ### What Proof Is - A collaborative document editor where humans and AI work as equals, suggesting edits, leaving comments, and tracking every character's provenance. ### Suggestions, Not Rewrites - Proof focuses on suggestions/edits rather than full rewrites to preserve the user's voice, using a track-changes model. ### Works With Any Agent - The tool is agent-agnostic, supporting Claude, ChatGPT, Cursor, Windsurf, and OpenClaw, with simple API calls for interaction. ### Provenance - Every operation tracks who wrote what, essential for maintaining human control and giving credit. ### Demo & Rollout - The team demonstrated adding an agent (R2-C2) via Slack and seeing its contributions appear live in the Proof document, highlighting the seamless, real-time co-authoring capability. ### Future Vision - Moving towards a more complex agent-native product where an agent can handle tasks like generating outlines or summarizing entire documents, leading to a more intuitive and powerful co-writing experience. ![Screenshot at 0:00: Dan Shiper introducing the episode topic on collaborative editing between humans and AI.](https://ss.rapidrecap.app/screens/P9XfYZh9CwA/00-00-00.jpg) ![Screenshot at 0:06: A screen share showing the 'Proof Launch – Podcast Agenda' document within the Proof editor interface, highlighting the collaboration status.](https://ss.rapidrecap.app/screens/P9XfYZh9CwA/00-00-06.jpg) ![Screenshot at 0:58: A sponsored content advertisement for Grammarly, contrasting it with the concept of human-agent collaboration.](https://ss.rapidrecap.app/screens/P9XfYZh9CwA/00-00-58.jpg) ![Screenshot at 2:19: The Proof landing page is shown, advertising it as 'Your agent's favorite way to write documents'.](https://ss.rapidrecap.app/screens/P9XfYZh9CwA/00-02-19.jpg)