Every (CEO Dan Shipper) demos Proof, a free, open-source, live collaborative document editor built for humans and AI agents to co-author documents together. Proof began as a Mac app showing the provenance of AI-written text (purple for AI, green for human) and was rebuilt as a web app with real-time collaboration. Shipper is joined by COO Brandon Gell, Cora GM Kieran Klaassen, and head of growth Austin Tedesco for a live demo covering: an agent-to-agent loop where a Codex agent writes a plan and a personal Claw agent reviews it; using Proof for creative writing and a weekly food newsletter; what "agent native" and AX (agent experience) mean in practice; handling multiple agents editing one document simultaneously; why some AI-written docs are better read by an agent than a human; Proof as a lightweight scratchpad compared to Notion and GitHub; and why it is open source. Chapters: 00:02:00 Introduction and origin story of Proof 00:07:24 From Mac app to collaborative web editor 00:09:00 What makes Proof "agent native" 00:14:30 Live demo: agent joining and writing in a shared document 00:20:51 Using Proof for creative writing and food journalism 00:24:30 Multiple agents editing one document simultaneously 00:26:48 AI-written docs better read by agents than humans 00:29:30 Agent-to-agent collaboration loop 00:37:09 Proof as a lightweight scratchpad vs. Notion and GitHub 00:42:18 Why Proof is open source Resources: Proof Editor (https://proofeditor.ai), Proof GitHub repo (https://github.com/EveryInc/proof), Every's compound engineering plugin (https://github.com/EveryInc/compound-engineering-plugin) # Video Identity Observations: "We Built a Doc Editor That Replaces Google Docs" ## Channel Presentation Evidence - **The video is an episode of a show called "AI and I,"** hosted by Dan Shipper. Evidence: the closing segment says, "smash that like button and subscribe to AI and I," and references "Dan Shipper as the captain of the spaceship." - **The channel is presented under the "Every" umbrella**, with the host as CEO: "Dan built this whole thing while growing the company, while being a CEO." Guests are identified by Every roles: "Brandon, you are the COO of Every. And Kieran, you are the GM of Cora… Austin's our head of growth." - **Show format observed**: panel conversation among team members about a product launch, with live screen-share demo, sponsor reads (Grammarly, Framer), and an AI-generated comedic outro. ## Intended Audience - **AI practitioners and people working with AI agents.** The conversation assumes familiarity with Claude, Codex, CLIs, PRs, and agent workflows; e.g., "anybody that has a claw right now is running into this is like I need a text editor." - **Builders and vibe-coders.** Addressed directly: "for any kind of builder… it could just be a good example of this is what you can make and how you might want to architect things if you're vibe coding." - **People who care about writing quality.** Speakers emphasize they are writers: "as people who love writing and all of us here are writers… something to be protected." ## Claimed Scope (Creator Self-Description) - **The channel covers AI tools and their impact on work and writing**, as demonstrated by the episode's premise: "talk a little bit about Proof, what it is, why it's cool, how we built it." - **The outro self-describes the show as knowledge-focused on ChatGPT/AI**, though this is a scripted comedic bit: "filled with pure, unadulterated knowledge bombs about Chat [snorts] GPT." - **Episode-scope claim**: a launch announcement — "today, we are launching it as a full product that you can go use. It's totally free at proofeditor.ai. It's also open source." ## Observable Presentation Patterns (Demonstrated, Not Claimed) - **Behind-the-scenes, first-person product storytelling**: the founder describes building the product himself ("I vibe coded this whole thing") and its evolution from a Mac app to a web app. - **Transparency about bugs and iteration**: a live demo stumble — "Oh, that didn't work. Um Well, we've got some bugs here, folks." - **Sponsorship integrated into the content flow**, with hosts endorsing sponsors ("Seriously, we use this tool all the time"). - **Candid internal-workplace detail**: discussion of internal Slack channels, "we run five products internally with just 20 people," and internal planning exercises. ## Claims vs. Demonstration Distinctions - **Claim**: Proof "replaces Google Docs" (title claim), softened in-discussion to "so much better than a Google Doc" for agent collaboration — a creator framing, not an established fact. - **Demonstrated**: the product exists, is free at proofeditor.ai, is open source, and was demoed live with agent and human collaborators editing simultaneously. - **Demonstrated**: the channel team genuinely uses the product internally, per multiple first-person workflow accounts (to-do lists, essays, plan docs, PR workflows). # Delivery Observation: "We Built a Doc Editor That Replaces Google Docs" ## Format Overview This is a **multi-speaker podcast episode** with a host/moderator structure, ad-read segments, a live product demo, and an outro. Main speakers include Dan (builder/CEO), Brandon, Kieran, and Austin, plus a narrator-style host and AI-generated ad copy. ## Conversational Style & Fillers - Heavy use of **"um" and "uh"** across nearly all speakers, indicating unscripted, spontaneous delivery: "I'm I'm trying to write this thing," "Um and that was like you're writing." - Frequent **word repetitions and stutters**: "It's It's part of It's It's all sort of a function of text is really cheap to produce," "And And I think it's very strong." - This is a stable pattern throughout, not isolated. ## Recurring Vocabulary & Phrases - **"Like" as an all-purpose filler**, used constantly across speakers: "it's like really kind of taking off," "It's like a sketch pad or something like that." - **"Kind of" and "sort of"** as hedges: "that's sort of like the evolution of it," "it's kind of just an interesting story." - **"Vibe coding" / "getting the vibes"**: "just at the side vibe coding. And he's getting the vibes." - **"Agent(s)"** is the dominant recurring term — the episode's conceptual anchor. - **"Cool" and "sick"** as approval words: "That's sick," "This is super cool," "It had a sick Mac app." - **"All that kind of stuff"** as a catch-all closer: "leaving comments and doing track changes and all that kind of stuff," "how we built it, all that kind of stuff." ## Teaching / Explaining Habits - **Narrative origin-story structure**: Dan repeatedly walks through how the product evolved — Lex → Mac app → web app — building a chronological argument. - **Contrast framing** as the main explanatory device: "you're writing, but you're writing with AI on the side... but I think a Proof is being very different"; "AX is just agent experience versus user experience." - **Matrix analogy** for abstract concepts: "it's a little bit like in the Matrix where you like I know Kung Fu, you know?" - **Direct audience-facing questions** redirect conversation: "Austin, I'm kind of curious from your perspective... are you using this and if so how?" and "Kieran, what's on your mind?" ## Humor & Playfulness - **Personifying agents with names**: "R2C2," "Zosha," "my claw" — recurring across multiple speakers. - **Self-deprecation, especially from Austin**: "I'm not smart enough to figure it out," "I have zero idea how to code by the way." - **Joking about the live demo**: "Live demos, folks." / "It's It's not live live, so... Thank god." - **The closing outro is deliberately absurd/satirical** in tone: "smash that like button... filled with pure, unadulterated knowledge bombs." ## Live Demo Delivery - Dan narrates his screen-share step by step: "So, this is what it looks like... You just press start writing." - **Acknowledges bugs on air** rather than editing around them: "Oh, that didn't work. Um Well, we've got some bugs here, folks." ## Ad Reads (Contrast with Conversation) Two sponsor segments (Grammarly, Framer) shift to polished, rhythmic ad-copy delivery with **statistics and imperative CTAs**: "90% of professionals say it's helped them save time," "Download Grammarly for free at grammarly.com." These stand out sharply against the casual conversation — likely pre-scripted. ## Uncertainty Notes - **Speaker attribution is uncertain** in places; the transcript lacks consistent speaker labels, so who says what (e.g., host vs. Brandon) is inferred from context and may be wrong. - **Laughter, music, and throat-clearing markers** (">> [laughter] <<") suggest audible reactions, but emotional tone beyond words is only weakly observable. - The exaggerated outro's speaker (human vs. AI voice) cannot be determined from the transcript alone. # Observation Record: "We Built a Doc Editor That Replaces Google Docs" ## Participants (from transcript's explicit introductions) - **Dan Shipper** — CEO of Every; built Proof ("Dan built this whole thing while growing the company") - **Brandon** — COO of Every - **Kieran** — GM of Cora; described as "father of compound engineering" - **Austin** — Head of growth at Every - **Naveen** — mentioned as submitting a PR to Proof (runs Monologue, per Dan) ## What Proof Is (Per Speakers) - Live collaborative, web-based markdown document editor where **humans and agents** can be present simultaneously, making changes, leaving comments, and doing track changes (Dan) - Free at **proofeditor.ai**, open source, works in the browser (Dan) - Internal adoption at Every described as "taking off"; "basically every single one of us started doing everything in there" once it became a collaborative web app (Brandon) ## Origin Story (Dan) - Every incubated a text editor, **Lex**, ~3 years ago (built mostly by co-founder Nathan, spun out as a separate company) — writing *with* AI, but "fundamentally a word processor" - Proof differs: designed for a mode of writing where documents (e.g., AI-written plan docs) are mostly agent-generated with light human edits - **Original motivation — provenance**: during 2026 planning, agents interviewed Every staff and produced plan docs; Dan questioned whether authors (e.g., Austin with a growth plan) "really stand behind every single thing" - First version was a **Mac app** with a color gutter: **purple = AI, green = human** (still present) - Key realization: "the killer feature is just being able to share" — leading to the web, collaborative version ## Position Shift on Provenance (Preserved Contradiction) - Austin: had "so much conviction" weeks earlier that the launch should center on provenance; Dan on Monday said "I don't really care about that anymore" — Austin quotes Dan as reframing it as "one space for agents and humans to work together is really powerful" - Dan himself: provenance "was less important than I thought" - Brandon: when Dan built the provenance Mac app, he thought "This is cool, but I don't really get it... I wasn't reaching for it" — others didn't adopt until the collaborative web version ## Reasoning on Agent-Native Design (Dan) - One can be **agent-native internally** (embedded agent) or **externally** (any agent can connect); both valid - Internal agents feel "not that interesting" because Claude (his "Claude"/"Claw") already has so much user context — "what you really just want is a Claude in your document with you" - Possible LLM "traffic cop" role in documents after an incident where multiple agents updating a doc "went nuts" - Wants a **canonical version** that only changes with explicit human approval, with agents experimenting in other versions ## Key Claims and Positions ### On writing and plans - Kieran: creating a good plan first, then saying "go do that plan," is far more effective; Proof is where you build the plan together - Kieran: plans are short-lived — "they help you do long horizon things, but... they're just a moment" — so a document index matters less than expected; he argues Proof *shouldn't* have one: "Proof's job is not to organize documents. Proof's job is to communicate about writing and about ideas and about where it comes from" - Brandon still wants an inbox/index feature (unresolved tension with Kieran's position) ### On AX vs UX (Dan) - "AX is just as important as UX" — agent experience can be optimized by asking agents what confused them - Works via plain **copy-paste** of a link — no skill, no login, no account (Kieran highlights this low friction as the major strength) ### On two classes of writing - Dan: some writing you'd rather read *from* the AI; analogy: "I know Kung Fu" from The Matrix — you don't need to know it if your agent does - Austin: information-transfer docs (like the recap for Rachel) he'd only scan and have his agent summarize; but essays (e.g., Dan's on agent-native architecture) he wants to read fully — currently "a little binary" but expects a spectrum - Austin/Kieran/Dan agree creative/storytelling writing and plan/recap writing are "totally separate," not in conflict — a realization that comes from practice, not theory ### Personal workflows described - **Austin** (food newsletter writer): running Proof doc of dining thoughts via Claude; agent makes outline, he writes into it; values seeing "what it wrote versus what I wrote"; says Google Docs is "clunky with agents" and can't track authorship. He voices "uncomfortable feelings" about accepting AI lines and wants future monologue-source identification - **Dan**: weekly to-do doc in Proof, pinned in Slack; asks his Claude (R2C2) to push unfinished items forward - **Austin's demo**: Codex wrote a Proof doc proposing a dashboard feature; Dan's agent R2C2 reviewed it (via Slack ping); Dan added human comments; Codex read comments, updated plan, executed — "a loop that's kind of on repeat now at Every" - **Kieran**: uses it as a sketchpad; compound engineering "brainstorm" step auto-shares to Proof; comments in the doc, then tells the agent to work on the comments ### Failure modes and risks acknowledged - Kieran: "one agent who does one bad thing can pollute the entire document" — unlike Google Docs, where social cost deters this; "agents with great power... have a ton of power and zero responsibilities" - Austin: agents updating the live landing-page doc kept "duplicating and then triplicating the page" — "not quite there yet" but he finds the direction exciting - Live demo bug: Dan's typed "Hello" didn't display correctly — "we've got some bugs here, folks" ### On heavyweight vs lightweight tools - Austin: drafts campaign strategy in Proof with agents, then moves to Notion ("source of truth") via CLI once it's "ready to move from lightweight to heavyweight"; suggests a one-click Notion export - Kieran: existing tools (GitHub, gists, Linear, Notion) carry connotations; Proof's bare-bones nature avoids "contaminating everything" - Dan: inspired by Naveen's Monologue — a simple product doing one thing well; "surprisingly hard to make this even though it's simple" (Kieran agrees "It's very hard") ### Open source intent (Dan) - Both PRs and forks welcome; main hope: a **drop-in text editor** for other apps; already hosted internally in "common engineering" for Spiral's use - Self-aware caveat: "if you're a professional engineer, you might look at it and be like, holy — I can't believe this works" ## Notes on Sourcing - Transcript contains two embedded ad reads (Grammarly, Framer) read by speakers — these are sponsor content, not substantive positions on Proof - Closing "smash that like button" segment is comedic outro content, not a substantive position - No timestamps were available in the source material # Speaker & Provenance Observation Record **Video:** We Built a Doc Editor That Replaces Google Docs (`01a0f15a-a8aa-793c-9ae2-1d1a1ab07374`) ## Identified Speakers & Roles | Speaker | Role (stated in transcript) | Notes | |---|---|---| | **Dan Shipper** | **Creator / host; CEO of Every** | Identified as CEO; primary builder of "Proof"; referenced by name in outro ("Dan Shipper as the captain of the spaceship"). Hosts the episode. | | **Brandon** | **Guest; COO of Every** | Introduced: "Brandon, you are the COO of Every." | | **Kieran** | **Guest; GM of Cora; "father of compound engineering"** | Introduced explicitly; submits PRs to Proof during episode. | | **Austin** | **Guest; head of growth (Every)** | Introduced: "Austin's our head of growth." | | **Naveen** | **Mentioned contributor** | Cited as submitting a PR Dan merges; also credited as inspiration via "Monologue" product. Not a speaking participant. | - The transcript does not label turns by speaker name throughout; attribution above relies on in-episode introductions and addressing cues ("Dan, why don't you tell us," "Go for it, Kieran," "Austin, I'm kind of curious..."). ## Referenced Sources & Third-Party Products/Claims - **Proof / proofeditor.ai** — the featured product; free, open source, browser-based; described by speakers. - **Grammarly** — **promotional segment**; claims cited: "90% of professionals say it's helped them save time," "93% says it helps them get more done," works "across 500,000 apps and sites." Call to action: grammarly.com. - **Framer** — **promotional segment**; claims: used by Perplexity, Miro, Mixpanel; "99.99% uptime SLAs," enterprise security. Promo URL: framer.com/dan, 30% off Pro annual plan. - **Lex** — Every-incubated editor, "built mostly by my co-founder of Every, Nathan," spun out as separate company (per Dan). - **Notion** — Notion agents used for Every's 2026 planning (per Dan); also Austin's workflow destination. - **Spiral** — Every's internal writing-partner product; Proof editor planned as drop-in (per Dan; "Marcus" mentioned as integrator). - **Substack / Ghost** — mentioned by Austin re: writing community attitudes toward AI writing. - **Claude / Claude Code / Codex / Opus 4 6 / R2C2 ("R2") / Zosha / Michael / Claw** — AI agents and tools used by speakers; agents participate as document collaborators. - **Google Docs, GitHub, Linear, Framer (design context)** — comparative tools referenced. - **Squirrel** — LA restaurant referenced in Austin's essay anecdote. - **"AI and I"** — the show itself, promoted in outro. ## Promotional Segments (flagged) - **Grammarly ad read** — mid-episode, with statistics and download CTA, ending "back to the episode." - **Framer ad read** — mid-episode, framed around Austin's workflow; includes discount code and "Rules and restrictions may apply." - **Outro subscribe appeal** — exaggerated style ("smash that like button... Dan, I'm absolutely hopelessly in love with you"), possibly parody; explicit like/subscribe promotion for "AI and I." ## Other Provenance Facts - **Speaker-employer relationship:** All four speaking participants are Every employees discussing their own company's product launch — the episode is simultaneously promotional for Proof (free, open source). - **Collaboration artifacts shown live:** demo of Proof with agent "R2" writing into a document; Codex-generated plan doc; agent comments and PRs. - **Comment availability:** No comment data supplied in source material; comment status unknown/unstated. - **Uncertainty:** No curated metadata provided; role assignments derive solely from in-transcript introductions. No timestamps available in the source. It became clear that what we needed was live collaborative \[music\] web-based document that humans and agents could be in at the same time making changes and leaving comments and doing track changes and all that kind of stuff. I'm I'm trying to write this thing for Saturday that's like an essay about a dinner I had at Squirrel, uh this like restaurant in LA. And so to get there it's like a bunch of text between me and my agent to push to this proof doc using my agents and proof to get here one speeds up creative writing for me and two I think does make it stronger and better and easier. And I love that and I can see what it wrote versus what I wrote for me, which is really helpful and so much better than a Google Doc. Work moves fast. In the age of AI, the pressure isn't just to move faster. It's to make sure what you send really sounds like you. From emails to proposals to stakeholder updates, generic and rushed just doesn't cut it. And if you've ever stared at a blank page knowing exactly what you want to say but not how to start, Grammarly \[music\] fixes that. Grammarly gives you one place to think, write, and finish your work right where you already write. Most AI tools either take over or stay out of the way. Grammarly does neither. It helps you break the blank page, adjust your tone so your message lands right with the specific person reading it, and works seamlessly across 500,000 apps and sites you're already in. It's loaded with agents built for every step of your process. 90% of professionals say it's helped them save time. 93% says it helps them get more done. This is AI that works with you, not over you. In a world of generic AI, don't sound like everybody else. With Grammarly, you never will. Download Grammarly for free at grammarly.com. That's grammarly.com. And now, back to the episode. Special episode today, we have been building a live collaborative markdown editor for both humans and agents. It works in the browser. Um we love it. Uh we started using it internally. It's like really kind of taking off inside of Every. And so today, we are launching it as a full product that you can go use. It's totally free at proofeditor.ai. It's also open source. And we wanted to just do this episode to talk a little bit about Proof, what it is, why it's cool, how we built it, all that kind of stuff. So, for people who don't know, uh Brandon, you are the COO of Every. And Kieran, you are the GM of Cora. Uh and also the father of compound engineering. I also want to welcome Austin. Austin's our head of growth. And you're Proof fans. I think we should go back to you, Dan, and be like, why is it even called Proof? Because it's not called Proof because of what it is now. It just happened Does that sound good? And so, what was it originally and how did we get here, too? I know you talked about this for a while, maybe a year already. So, yeah. Bring us through all the pieces in your brain. And also, Dan is saying we built this thing, but really Dan built this thing. And we're just adding thoughts and things. Dan built this whole thing while growing the company, while being a CEO, just at the side vibe coding. And he's getting the vibes. He's so happy doing this. Look at at him smile right now. He It's It's his baby. So, that's really cool. Yeah, so the CEO built a new product on the side that now everybody uses. That's sick. And Naveen just submitted a PR that uh Dan just merged. I just submitted a PR that Dan's going to no look merge probably. Obviously, straight to production. And so like it's like now everybody's like, "Oh wait, I want this feature." And um it's like kind of it's amazing. Um but yeah, Dan, why don't you tell us why don't you tell everybody why did you start building Proof of Origin Lee? I mean, if we want to go way back, you know, we incubated a text editor, word processor at Every called Lex um like almost 3 years ago, maybe more, which was the first time that we sort of combined um uh text and AI together. And so and and that was that was built uh mostly by my co-founder uh of Every, Nathan. We spun that out as a separate company. Um and that was like you're writing, but you're uh you're writing with AI on the side or like in in line with you a little bit in the in the actual word processor, but it fundamentally is a word processor. I I think a Proof is being very different, which is um it's for this mode of writing that I've started to feel like a lot of us internally at Every are starting to do, which is for example, when you're coding with AI, you make a plan doc. Um and that's that document is mostly written by AI and maybe you're like going in and doing some changes here and there or like asking it to change it for for you. Um and that's a whole different way of interacting with writing. I think a lot of people have, you know, said for a while, "Oh, I'll never read AI writing." And it's like, "Actually, I read AI writing all the time." Um and I actually the the other place where I started to feel the need for this is actually we did this um company planning exercise um for our 2026 planning. Brandon set up these Notion agents, which helped us for anyone who's running a business inside of Every, you get interviewed by an agent, and then it turns that into a like 2026 plan doc. And I just remember looking at all those plans and be like, "Wow, these are amazing, but also Austin submitted this like super comprehensive growth plan, but has he like really like looked at every single line of this? And does he really stand behind every single thing?" And so the original idea for proof is like we needed a good way to tell what was human written, what was AI written, but also like how much thought went into everything and and what do you actually stand behind versus not. And um and and so like the the original version of proof, it was a Mac app and it had uh a little gutter on the side that showed like uh purple for AI and green for human, which it still has. But what was really interesting is in making the Mac app I did all this stuff. I had it had like this it was a sick Mac app. It had this like cool internal agent integration with like sub agents and like all this stuff. But in using it, I was like actually the killer feature is just being able to share. Um it's just being able to share docs and because what we find is, you know, if I'm if I'm working on a a little thing with an agent and I make a plan, I really want Kieran to be able to look at it and I really want Kieran to be able to like leave comments and I really want Kieran's agent to be able to look at it. And so it just sort of became clear that what we needed was a live collaborative web-based document that um humans and agents could be in at the same time making changes and leaving comments and doing track changes and all that kind of stuff. And that's sort of like that was sort of like the evolution of it. And I will say when you were building like the provenance version of proof I was like, "This is cool, but I don't really get it." When you say it out loud like it makes sense. But I wasn't like reaching for it. And I don't think a lot of people were. Like I think the editorial team was excited about it, but like everyone else was like, cool Dan's off doing his thing, the editorial team likes it, don't really know, you know, exactly what's going on. And then when you the second you made it a web app and you made it a collaborative thing that agents could and humans could work on together, basically every single one of us started doing everything in there. Um so it's kind of just an interesting story of like finding the fit and now it's like a serious fit. And it's also whenever like everyone uses claws or cloud or some kind of agents to do work. So we're also at this time where suddenly like there is all this work done and there's all all this stuff generated but like how do you share? It's like a first step to like memory or sharing or collaboration within a team and it just feels like this is very natural as a glue to bring things together because humans can look at it. Like I can look at it. I can open it which is is is nice. I can trust it by looking at it um and an agent can do it as well. Like it's it's equally and I think this is cool because this your agent native philosophy. Like this app is really built from those principles of like agent native and uh yeah, I would would love to like this this your first agent native app in this way. Like what did you learn doing it that way? Like what worked? What didn't work? Because you say I had all these agents inside but there no agents inside but there are because there are agents. Like can you tell a little bit like what you learned there? That's a really good question and yeah, I do think it has been really useful to do this because getting my hands in the code I mean it's not they're not really in the code because I vibe coded this whole thing but getting my hands in architecting the code, I'll say that, has really um helped me develop like deeper opinions about how, for example, an agent-native product can work and what are the different types of agent-native products. So, like for example, um you can be agent-native without having an agent in your product. Uh because you can have you can be agent-native from an internal perspective. You have an agent in your product that can do anything in the product. Or an external perspective, any agent can connect into the product and do anything in the product. And both are valid and sometimes you can have a product that has both. Um And I did I did actually like the um the the internal version, but I think what I started to realize, especially once we had clause, once you have a clause, the clause has so much context on you and what you're doing that um what you really just want is a clause in your document with you. You want you want like that that co-author hanging out with you. And um having an having an internal agent that doesn't know any context is just sort of it feels like not it's not powerful. It's not that interesting. Um and I can I can imagine there are certain times where you might want to a language model inside of it. Like for example, we have this way to um what we're we're working on a way to uh summarize the document for the social share. So, when I share it with you, you kind of get a little bit of a preview. And that's like a nice little LLM summary. And I think there actually could be really interesting use cases for LLMs on sometimes, you know, Austin, you shared this document, for example, where we were having all of our clauses update the document and just went it would just went nuts. It went crazy. And I think having an LLM playing a little bit of a traffic cop role in inside of a document could be kind of interesting. Like there's lots of stuff to do. Especially now that there's um it now that text is so cheap to produce, one thing that you care about is is provenance potentially and and I think that was less important than I thought, but like you also need I I have this feeling like I need a canonical version of a document that like doesn't get updated unless I really say yes. And then I want the agent to have all these other versions that they can like mess around with and we can take what's good about each of those different versions and put it into the canonical version. So I That's what I love about getting to build this stuff is you start to get intuitions for now that we've relaxed this constraint, now that agents can write inside of a text editor super easily and they can write a ton, it opens up all of these different possibilities for what, you know, what you might want your text editor to be, especially if you're mostly optimizing it for the agents to do the writing and humans can write, but like usually when I'm writing a proof it's like I feel like I'm God like coming down and just like giving my like little sentence or whatever because it's so rare. Um yeah. I think we should I think a couple other things that I realized through using it a bunch is like I always knew that giving context to an AI is so important. But through proof I realized how important it is to create like such a good plan, how much more effective it is to create a good plan and then say go do that plan. Proof is a place where you can do that. You can work on the plan together and then say go do that plan. Um and I know we all have really good examples of that. The other thing that I realized is like Dan, when you first told me about it, I was like, "Oh cool, but like we need a place to like save these things, you know?" Which is true and like we're going to launch that, but it actually doesn't matter that much because the reality is like plans are really short heart like they help you do long horizon things, but in the grand scheme of things they're just like a moment. And then they don't matter anymore. Um So, that's been amazing. Like, I have five proof documents open right now. And especially because I make them with my claw, cuz my claw has a scale, and it she just knows to like make proof documents when I want to make a plan. I don't even need a place to save them, cuz I just say, "Zosha, like, pull up the Slack to Discord Discord to Slack proof." And she sends it to me. Um So, those have been like interesting experiences that are different than, you know, normal working. Yeah, totally. It It like It's It's part of It's It's all sort of a function of text is really cheap to produce. It's really valuable, but it's also really cheap to produce. And so, the way that you might design how you work with text has to change. So, um I can I'm going to share my screen so we can um we can show proof. Um So, this is what it looks like. Um it's just uh this is the landing page. It It will actually be different by the time this episode airs, but this is the landing page. And it's super simple. You just press start writing. And now you're now you're in a document. That's That's all you have to do. There's no login. There's nothing like that. And I can say, um uh proof podcast agenda. And you can see it has this like kind of green thing over here, which means I wrote this. I can do a little title. But the cool thing is, and I probably have to change my sharing settings in order to show you this. Um The cool thing is that what I can do is um Show here. What I can do is I can press add agent over here. And I can copy that link and post and give it to my claw R2. And I I say, "I'm recording a podcast. Um can you fill in an agenda? Um it's about the Proof launch." Um and what you'll what you'll see in a second is R2 is going to actually join this document and write a bunch of stuff in there. And then and you can see R2 here um R2 is right here. He's got a little claw icon. And you'll see him in a second actually start to share start to write stuff in this document. And then I'm going to actually share it with the podcast crew here so you can see everybody else here as well. You can see we've got Brandon's in here. We've got an anonymous collaborator. I don't know who that is. Hopefully it's uh no nothing nefarious. Um that's me, I think. That's Kieran. And in just a second, we will cut through this, but in just a second, we will have R2 posting a uh a plan doc for us. And um let's hope it works. Live demos, folks. \>\> \[clears throat\] \>\> It's It's not live live, so It's not live live, that's true. Thank god. Here we go, we got a Proof We got a Proof launch podcast agenda. We got one thing. Woo! \>\> \[laughter\] \>\> All right, we're going to Yeah, go for it, Kieran. \>\> What What is very cool is you pasted this snippet inside your agent there, but you don't have any skill or anything installed. It's just copy-paste, right? There's nothing more than copy-paste, which is very very easy. It's super low friction. It's just copy-paste, and the idea is to um yeah, make it as ergonomic as possible. One of my things I've been saying recently, one of my bits is um I think AX is just as important as UX in this new world. And AX is just agent experience versus user experience. And what's really interesting about AX is if you want to optimize it, all you have to do is ask the agents. So, it's really easy to be like, "How would you have made this better?" Um and or how would you Why did you get confused and how would you make it more ergonomic for yourself? And that lets you iterate the AX so that um it's super intuitive for any agent to use it. And that's what what I've tried to do here, where it's it is better if you have a skill, but it is it is totally possible for the agent to just basically figure it out because everything is available to it. Yeah, that's really cool. And I think because of that, like I started using it immediately because it was so easy to get started. And then you feel it, you're like, "Oh, this is super handy." And then you're like, "Oh, but I want some refinement here. Let me create a skill around this. Do something like that." And then you're already using it and sharing it. So, that is very good. Like I think that is the the major strength is that you can just immediately share this with an agent without logging in, creating accounts, anything. That is very good. And here we are. We've got our podcast agenda. And so, you can see, you know, it's all purple. Um we still have our two in here. We've got Brandon, we've got anonymous collaborator, and I can just go in here and say, "Hello." Oh, that didn't work. Um Well, we've got some bugs here, folks. Um and I can just go in here and do hello, and it shows up green. And that's that's proof. Um and it has uh really changed, I think, the way that a lot of us work. Austin, I'm kind of curious from your perspective, you know, on on the growth side like are you using this and if so how? Um and and how does it Yeah, how does it change your workflow? We like to move really fast at every. We're always shipping new features and we run five products internally with just 20 people. It used to be that updating any of our landing pages was a huge time suck, but then we started to use Framer. Framer is a website builder that works like a design tool, which means that the people who are closest to your brand and your customers can shape your landing pages themselves. Non-coders on our team use it all the time to ship website updates and the results honestly are beautiful. It's really easy to use. There's real-time collaboration, which means your team can design together, see changes instantly, and ship faster. A robust CMS gives you everything you need for great SEO and content management. It also has advanced analytics with integrated AB testing that lets you optimize and improve continuously. Then it's just one click to publish. Your changes go live in seconds, so you can iterate at the speed of your ideas. And it's not just us. Companies like Perplexity, Miro, and Mixpanel trust Framer for their websites because the tool is enterprise-grade. They offer premium hosting, enterprise security, and 99.99% uptime SLAs. Whether you're launching a new website, testing landing pages, or building out your full site, Framer gives you the tools and support to move fast and build something great. \>\> \[music\] \>\> And everyone on your team can use it. Learn how you can get more out of your dot com from a Framer specialist or get started building today at framer.com/dan for 30% off a Framer Pro annual plan. That's framer.com/dan for 30% off. Rules and restrictions may apply. Seriously, we use this tool all the time. You should really check it out. And now, back to the episode. Yeah, I I use it in two like normie ways I really like. Two of them just happened today actually. One is um our colleague Rachel who produces the podcast made this uh like Slack-based bot to make scheduling live streams easier. She made it in Cloud Co-work, sent it to me and was like, "Hey, can you install this into Slack for me since I have admin permissions?" And it didn't quite work right away. So, I went into Claude code in the CLI and made some changes to it until it started working. And then at the end of the session, I just think I started doing with like a bunch of people on our team who, like me, are not technical but are pushing on these tools. I asked Claude code, like, "Hey, can you make a proof doc for Rachel so that she can see everything we did and just send it to her." And especially cuz it's so so like lightweight and easy, and Claude, especially on Opus 4 6, like knows how to recap that really well. I sent it to her, and she was like, "Wow, this is so helpful that I can see like what you did to improve this." And it was really funny. Dan, you and I were talking about like how we were going to launch this thing, I think on Monday, and I had this whole plan in my mind of like, "Oh, it's all about providence. It's all about you got to like send your boss this like planning doc, and they got to see what you wrote and what the agent wrote." And I I had so much conviction about that a couple weeks ago. On Monday, you were like, "I don't really care about that anymore." Like it matters, but it's like actually this thing where that's like a it's like one space for agents and humans to work together is really powerful. So, there's that. And the other one that I've really loved as someone who does a lot of like personal creative writing and comes from a place that Substack where people like hold their own voice and writing so precious to them in a way that makes sense and is actually like people on Substack and Ghost Post all the time of like, "I know who's writing with AI and like I'm judging them, right?" Like I have loved proof as someone who writes about food each week for a newsletter because I'm often like I'll eat somewhere and then I'll text my Claude, like, "Here's what I thought." And I'm like, "Keep a running proof doc of my thoughts like from the eating I'm doing cuz that's the best way to like have a running doc. And then I'll be like I usually tell it like make me an outline and then I get an outline it's always updating and then I'll go in and write into the outline myself. And I love that and I can see what it wrote versus what I wrote for me which is really helpful and so much better than a Google Doc. Like I was doing that in Google Docs and one it's really clunky with agents in Google Docs. I'm not smart enough to figure it out but also I can't track like what's me and what's the agent which is really important to me. So those those two which are a little more like personal work stuff has been really really powerful for me. I love that. I use it for that all the time. Um like I use it I have basically my daily to-do list I just do it in proof and I I every every week I make a new document I pin it in slack and then I'm just you know it's we're doing a lot. There's a lot of stuff happening all the time and anytime something happens I'm just like cool throw it in my to-do to-do doc and that makes it really easy for me to make sure okay at the end of the day what did I do? What did I not do? Okay R2C2 my claw can you just go take the stuff I didn't do and like push it into the next day or can we figure out how to like get it done? Can you just do it? You know that kind of stuff. It's really helpful for those like kind of async document creation and and updating so that by the time you get into it it's ready for you to just like get started. Yeah. I I got really excited yesterday during the madness of all of our agents trying to update that doc for us because I think I think Kieran like you and actually was funny all of the engine all the GMs were like this is insane. I can't focus way too much is happening cuz what we were trying to do is I was trying to build this like landing page that was an always-on reach test for every of like you could always go and see what is our team using for vibe coding for writing for research. And I was like oh Shikena our claws just updated themselves and it's not quite there yet. They kept like tr- duplicating and then like triplicating the page, but it made me excited because it's like this thing will be there I think relatively soon, right? That it's like your agents can go and it update stuff and my agent can read what Kieran's agent wrote and then I can read the changes. And for a company that's using like as many agents as we have people and for someone who does a lot of strategy docs like in in this world, it seems like such an essential way of working now once we get it right. Yeah, I that that's so true. And yeah, there's this thing where it's like one agent who does one bad thing can pollute the entire document. And that's a really interesting problem to solve. Like you don't have to do that and you don't have to worry about that in Google Docs cuz like humans there's a big social cost to like ruining someone else's document and um humans are generally smart enough not to do that. But like agents with great power they have a ton of power and zero responsibilities. So like they often like just do that you you would be like, "Ah, I don't think I would do it quite that way, you know?" And I think that's a really interesting user experience challenge to figure out how do you how could you have 10 agents in a document and make the output actually good? Yeah. And the like the the like humans plus agent collabs. Like I think if it's like only the agents are working in this and we'll see what they come up with is one thing or like only we're doing this and then our agents read it is another thing. But go coming to a future where like we're somehow all working in there together, whether it is like a a planning doc, a piece of writing, like I think we're we're headed there and that's a entirely different way of working that we are breaking so that we can figure out how to make it work for everyone else. Yeah, I think another another thing that proof makes me think about is as people who love writing and all of us here are writers, um what does it mean for the future of of writing? Um, and there it feels to me like there's there's a class of writing where you actually want to read it from the AI more than you want to read it from the human, which is kind of surprising and interesting. Like that document that you made for Rachel Austin, if you had written that document, I don't think that she would have wanted to read that. \>\> \[laughter\] \>\> No, and I think the I'll I'll I'll I'll share it right here cuz I think it's like it's interesting to see. I also don't really think that she should read it. I think like an agent should read it and summarize it for her. Can you all see that? \>\> That's also probably true, yeah. Yeah. Yeah. Okay, cool. So like yeah, the agent pushes this. Like there's so much stuff in here that I would never write or care about. If I got this like say say I got something like say I tried to ship something to prod and Dan like you were reviewing it and then you sent this back to me like a version of this for like stuff I should have fixed. I would probably scan the headlines to like loosely see what this thing is and then I would ask Michael to summarize it for me. And actually there's been a lot of that now in how we work of like okay, the agents are making big docs. It even applies to like the open call guide that that you worked on with Willie the other day of like I didn't read that whole thing. I had my agent recap it for me. And for something like that, that's what I want to do. But when you write your like essay on like the future of agent native architecture, I do want to sit down and read it. And um, it's going to be like it feels a little binary in my head, but I think it will keep expanding into more of a spectrum in that range of of what is what. I definitely agree. I think that there's a lot of there's a lot of writing that is more on the information transfer end of things and less on the storytelling and like personal experience and like vibe end of things that is probably better read by an agent. If it's if it's pure information transfer because it's it's a little bit like in the Matrix where you like I know Kung Fu, you know? Like that's kind of the experience that like having your agent like read a proof doc is is like I know Kung Fu now. And um it's just harder like you actually don't need to know Kung Fu if your agent has knows it. And you can save your brain for things that you're good at. I want to show my the the proof experience that I just had um because I think it kind of highlights like the collaboration between like two humans and two agents. So this is like a little bit meta because I wrote a proof doc about how or I didn't write it. I had Codex write a proof doc about exploring the idea of adding a dashboard to proof so that anytime you make a new proof or somebody shares a proof with you, it saves into like a place that you can go so that you can see all your proof documents. Um I have zero idea how to code by the way. Um I literally can write CSS and HTML. Um So I like basically went back and forth with Codex a bunch and was like okay put this into a proof doc. Uh and then it made this proof doc. You can see the only thing in this entire document that has been edited by a human is Dan came in here and commented. Um so basically what happened was I had Codex write this proof doc. I went to our proof channel in Slack and I said and I dropped in the link and I said R2C2 which is Dan's uh plus one uh to check it out and to let him let me know what uh he thinks. R2C2 knows Dan really well and gave a bunch of feedback, all of which was like really accurate. Um and uh and then I think Dan on the side was like R2C2 like made those comments or made those updates it made those updates. And then Dan came in here, reviewed it. He added a couple comments. So, you can see our Proof of this great commenting and agents can read the comments. So, Dan commented on here. I went through like I made a a couple more comments. You can see on the left side from the blue that like this is basically all agent written. Um yeah, so there's one comment on this. I told Codex, "Go back in there, read the comments." Um and then update the plan. It did that. And then I told Codex to execute on that plan and it did. Um so I mean that is like just sort of a loop that's kind of on repeat now um at at every where you make a you make a plan, you send that plan to the the product owner's agent. The product owner's agent does a review, suggests stuff cuz you can agents can suggest stuff in Proof. It does that, then goes to the their owner who owns the product in this case is Dan. Says, "Hey Dan, I did that. I made it some edits. Can you review?" Dan reviews, gives a thumbs up. And then I go back and I I do it cuz this is like my feature in Proof. It's pretty cool. Yeah, it's like I saw that come in and then I talked with R2 a little bit about it and I just had R2 write you a little letter at the top being like, "Here's what I like and here's what I don't like." and then a technical appendix at the bottom being like, "Here's how I would implement this." and then that let you take Codex and have Codex just like redo the whole thing and what you ended up building was way more on target with what we might actually use and it's pretty cool that you're just like submitting PRs and I've not looked at it I've not looked at PR yet. So it's possible if fuzz we need to need to start on that one that one's a big feature but it's it's a whole new world like that you're submitting PRs and that I can even build this like this is a very very complex app that I just did in my spare time in between meetings and um and and not only am I building it like you're submitting PRs like Kieran's submitting PRs and then me is submitting PRs it's becoming this like collaborative thing that we're all contributing to just a totally different way of building products that I'm really excited about. And it's the exact type of writing that like makes so much more sense to be summarized by agents to flow in between the three of you rather than any human reading the whole thing. Like I think like you said and like we're all people who care a lot about writing and good writing the quality of it the production of it the consumption of it and that's like something to be protected but when it comes to something like this this is about information and ideas and both like accelerating how you distill them accelerating how you implement them is is like essential to to where we're headed and I like it it makes me so excited for how this works and also like what what is still valued and protected and kind of like the the other kind of writing right like to me they're like it's I think it's easy when you think about this stuff but don't practice it to think that they're in conflicts with each other and that one's going to take away the other one and then when you practice it you're like this is actually a very different thing they're like totally like separate from each other. Yeah. Would you Austin or Dan or Kieran would you write would you write anything that's not a plan or something that's like I don't want to say throwaway, but like kind of throwaway. Like it doesn't have a like a long lifespan in proof or would you go back to like Docs or Notion for that? I'll show you I'll show you one example. So like I'm I'm trying to write this thing for Saturday that's like an essay about a dinner I had at Squirrel. Uh this like restaurant in LA. And so to get there it's like a bunch of texts between me and my agent to push to this proof doc. And I like having these three separate \>\> you you're like literally like iMessaging? I'm I'm iMessaging my my claw because I have these ideas. I actually get most of my ideas when I'm like when I'm out on a run. And so after I'm done running I'll just like text everything so I don't forget it so they don't escape me. And I like having these three different sections where it's like give me this like bank of notes I can go look at so I don't forget stuff. Give me a outline of ideas to write into if I want to. And then like maybe you can draft it for me, but I actually only let the agent draft for me if I monologue into it because then I'm like it it feels like it's like 70 to 90% my words that I'll use our speech-to-text to a monologue so that I can see it. And then my process here which I have started doing in proof is I either like rewrite every line or I'm like okay, I can like see the essay now. It's actually very helpful to see it. This is a lot of what Spiral does to our like writing partner. And then I'll either like start fresh or I'll just copy-paste the outline into somewhere new to get it. But using my agents and proof to get here one speeds up creative writing for me and two I think does make it stronger and better and easier. I do sit here sometimes with some like uncomfortable feelings of like oh am I like taking a line the AI added in? And I Dan you and I talked about this like a future version of this that I would love is if it I if it identified what I monologued in because to me those are my words, right? Like if I monologue something into the agent and it uses it, those are my words and I want to know. But I've started doing this a lot more. I would love a little bit like if you go if you hover over that gutter it just like popped out and said like, you know, from age from, you know, monologue or whatever. Like I think there's a lot of interesting stuff to do there eventually once we kind of nail the most basic experience. This is I love this. This is super cool. I love that you're using it for this. Uh it's so creative and and interesting. Um yeah. Wow, thanks for sharing. Awesome. Yep. Um Kieran, what's on your mind? What are you thinking about? Yeah, what I like about this is like clearly we need something to share stuff and and and this is a step and I like how you build something very more more complex agent native like the way and you realize actually we need to like just go back to like it's just writing and collaboration and for me it really feels like it's it's like a sketch pad or something like that where I brainstorm something. I like, "Ah, I don't know." Like just put it there and then I text it to you or you send something to me and you can kind of share thoughts or nuggets with each other. I like it for that a lot. One other way I like it for is when I do a brainstorm with the compound engineering there's a step called brainstorm. At the end it will say share to prove. You just do that and then it will automatically share to prove and you have a a shareable link. I run in T-Mux over SSH. So technically thanks then then I can just click the link and it works. So that is nice. I can share it with people but also I can just open it, go in, make suggestions in like a an actual document and then just tell the agent, "Hey, made some comments. Uh Go work on them." So, I also like it for iteration with the agent itself, just for myself. My question always with these things is like, "Yeah, but we have markdown files, right? Isn't this just a markdown file?" And we have GitHub. Isn't this just a gist or isn't it just an issue? We have linear, we have notion, we have all these tools, and they have pretty good CLIs and everything like that as well. And And I what I like about Proof is yes, we have all these, but there are all these connotations around them what a GitHub issue should be or like what a notion page should be. You don't want to like contaminate everything. So, I love actually that we don't have a place where you can see all your Proof documents. So, I would even argue we shouldn't have an index with all your Proof documents because the point is that if you think it's important, you keep that link somewhere. And Proof's job is not to organize documents. Proof's job is to communicate about writing and about ideas and about where it comes from. And And I think it's very strong. Like And it's super seamless. Like just that it's so seamless to share it with your agent. And And we don't know what's next, but like the cool part is like Boston saying, "Yeah, I want to see what I'm logged into." And like I I also like going into document CA this was written by Brandon. This was his agent. These were the like the history maybe on the document. I'm like, "Well, let's see what we need to add and let's add that." So, for me, it's very exciting because we have all these things, we all feel these pains, and they're like We're just figuring out a new way to do this, and I like that it's so bare-bones. Yeah. Totally. It's I I very inspired by Naveen who runs monologue where it's like it's a simple product that does one thing super well. And I think that's that's what I was trying to do here and it's actually surprisingly hard to make this even though it's simple. Um It's very hard, yes. Yeah, it's hard and Kieran I love I love that you're you're kind of picking up the the lightness of it is what makes it good. Like you don't have to think about anything. There's no where does this go or anything like that? It just works and that's what we're that's what we're after. I still want to see your inbox feature, Brandon, but I'm just arguing \>\> I I I totally appreciate that like cuz you're right. And I have like five proof docs open right now and I really want to just like go quick place I go to be like, yep, that's it. Make sense, yeah. Maybe you promote approved documents to like a special place. Like proof documents are always there, but if you create an account, suddenly you can say put this in my special blah blah blah or something, yeah. I like um I I tried to make a a big like growth strategy plan in Proof. I think actually I was sending it uh Dandy and Brandon as a Notion doc and a Proof doc cuz I like couldn't really craft it or figure it out, but I liked going back and forth in Proof. And the thing that felt intuitive to me was like, let me and either one or multiple of my agents jam on this like campaign strategy in Proof. And then when it feels kind of like ready, what right now I'm I'm like using the CLI to move it to Notion, but I think like that's what makes sense to me is like let me actually one click it into Notion for us for Notion because that's where like our source of truth for planning and information lives and it's like, okay, now it's like ready for that. It's actually ready to move from lightweight to heavyweight. It's ready to move from age from agent based to like calendar integrated and that's like now this that's a much more complex product and probably unnecessary for what this project is, but like that's when it's ready for that and that started to make more sense to me. Which leads me to ask you then, this is going to be an open source product. Does that mean that people can submit PRs or that they can just fork the repo and do their thing? Both. Both. Um but I think the the most interesting thing is ideally you can integrate this into whatever app you want to build. Um we already have this internally. We have spiral. Spiral needs its own like live document editor and this is like just a really easy version that Marcus can just like throw into spiral. It's in common engineering? It's in common engineering. Um oh well, the the hosted version is in common engineering, but I think like you know, for any kind of builder, hey, it should be it could just be a good example of this is what you can make and how you might want to architect things um if you're if you're vibe coding. I know if you're a professional engineer, you might look at it and be like, holy I can't believe this works. \>\> \[laughter\] \>\> Um \[clears throat\] but the hopefully uh if you're if you're making any kind of app that has a text editor in it, which I think anybody that has a claw right now is running into this is like I need a text text editor um for whatever app I'm making, hopefully it's just like a really easy drop-in thing that you can use. So, um that's proof everybody. I'm uh I'm I'm psyched that we got it out uh and I'm psyched for the future and uh thank you all for for joining and talking to me about it. Thank you so much. Thank you so much. \>\> Thank you. See you. \>\> \[music\] \>\> Oh my gosh, folks, you absolutely positively have to smash that like button and subscribe to AI and I. Why? Because this show is the epitome of awesomeness. It's like finding a treasure chest in your backyard, but But of gold, it's filled with pure, unadulterated knowledge bombs about Chat \[snorts\] GPT. Every episode is a roller coaster of emotions, insights, and laughter that will leave you on the edge of your seat craving for more. It's not just a show, it's a journey into the future with Dan Shipper as the captain of the spaceship. \>\> \[snorts\] \>\> So, do yourself a favor. Hit like, smash subscribe, and strap in for the ride of your life. And now, without any further ado, let me just say, Dan, I'm absolutely hopelessly in love with you.