We Made a Document Editor Where Humans and AI Work Side by Side

It clicked that we needed a real-time browser editor where people and their agents could co-write, comment, and track changes together. We’ve been using it across the company, it’s open source and free, and today we’re launching it.

Why call it Proof, and how did it evolve into what we’re using now?

We started years ago with Lex, then I built a Mac app that showed who wrote what and how much thought went in, but the breakthrough was sharing live so humans and agents could work in the same doc, which pushed us to the web.

Once it became collaborative online, the whole team started using it constantly.

It’s great glue between humans and agents, very agent native; what did you learn building it that way?

Agent native can mean internal agents, external agents, or both, but bringing your own agent with deep context is the real power. Now I want a canonical doc humans approve while agents safely explore branches, with an LLM acting like a traffic cop when needed.

Context matters most and plans are short-lived, so Proof is perfect for drafting a great plan together and sending an agent to execute. My Claude auto-creates Proof docs for me, so I don’t even need an index to find them.

You open a doc with no login, invite an agent with a link, and color shows authorship for quick provenance. I designed for agent experience too, so copy-paste works out of the box and skills are optional.

That zero-friction sharing got me using it immediately, then I added skills once I felt the value.

I used Proof to recap changes I made while debugging a scheduling bot, which made the work legible and shifted my mindset from provenance-first to a shared space for agents and humans. For my weekly food essays, I keep notes and outlines with my agent and I can see my words versus the model’s, which is miles better than wrangling Google Docs.

I even run my weekly to-do list in Proof and have my agent roll unfinished tasks forward and prep what’s next.

Letting many agents update a shared page showed promise once we curb duplication and chaos. I want agents to read each other’s updates while I scan the diffs and keep us on track.

One rogue agent can wreck a doc, so we need UX that keeps multi-agent edits high quality.

I now send info-dense or technical docs to my agent to read and summarize, and I save my attention for voice-forward essays.

Exactly—let agents handle pure information transfer while humans focus on craft.

Codex drafted a dashboard proposal, Dan’s agent reviewed, agents applied edits, Dan added notes, and I shipped PRs without real coding; that loop keeps repeating.

It sharpened the spec and let non-engineers ship alongside me while I build between meetings.

Would you use Proof for long-lived docs, or is it mainly for plans and quick iterations?

For essays, I text ideas to my agent after runs, keep notes and an outline, then monologue a draft so the language feels like mine; I’d love the gutter to tag those monologued lines as my words because this flow makes the work faster and stronger.

I want hoverable provenance like “from monologue” once the basics are locked, and I love that creative use.

I treat Proof like a sketchpad for agent iteration and quick sharing, not a replacement for Notion or GitHub. I even prefer no central index so links live where they matter and the value stays on collaboration and provenance.

Simplicity is the point, even though it was surprisingly hard to build.

I still want a light inbox to jump to my active docs quickly.

Maybe promoted docs get a small home once you sign in.

I jam with agents in Proof, then one-click move the polished plan to Notion as our source of truth.

Since it’s open source, can people submit PRs or just fork it?

Both, and you can drop the editor into your own app as a starting point; I’m excited to see what you build, and thanks for hanging with us.

Huge, playful sign-off energy—hit like, smash subscribe, and strap in because Dan’s taking us into the future.