Interview (Ep. 39, hosted by Dan Shipper) with Vicente Silveira, cofounder and CEO of AI PDF, a GPT-wrapper tool for summarizing, chatting with, and organizing PDF files. Despite predictions that ChatGPT's native PDF support would kill wrapper products, AI PDF reached roughly 500,000 registered users and nearly 3,000 paying subscribers with a five-person team funded by a friends-and-family round. Topics covered: the advantages of staying small and specialized, building with early adopters, why lean startups may out-innovate frontier AI companies, when AI startups should raise venture capital, and the emerging role of "AI managers" overseeing AI agents. Includes a live demo of an agent integrated into AI PDF. # Channel Identity Observations: "AI and I" (Ep. 39 with Vicente Silveira) ## Explicit Self-Description Evidence ### Host and Show - The host identifies himself as **Dan Shipper**, addressed directly by the guest: "Sante, welcome to the show. Hey Dan, thank you for having me here." (Guest later confirms: "it's a journey into the future with Dan Shipper as the captain of the spaceship.") - The show is named **"AI and I"**, per the closing segment: "you absolutely positively have to smash that like button and subscribe to AI and I." - Dan describes himself as running a media company: "I run a media company and um I employ a lot of people, like, you know, editors and writers... and designers." - He references his organization **Every** repeatedly: "there's a lot of overlaps between the kinds of things that you're building and the kinds of things we're building internally at Every" and "we have a product called Spiral." ### Host's Stated Editorial Intent - Dan states the show's purpose in his own words: "one of the things that I would love to do with this show is show people the easy sort of ways that they can get started and... what the most interesting or smartest or furthest ahead people are doing so that we can like bring forward everyone else to use it." - He frames the mission as "raising the floor... creates a lot more economic opportunities for people." ## Claimed Scope - **Topic scope**: Conversations with AI builders and founders about building in the AI economy — this episode covers the guest's product (AI PDF / AI Drive), competing with large AI platforms, lean-team startup economics, fundraising, and AI agents. - **Format demonstrated**: Long-form interview (~one hour), with **live product demos** where the guest shares a screen and demonstrates agent workflows in real time ("This is a live demo here. So, anything can happen"). - **Guest type**: AI startup founders/CEOs — the guest is introduced as "the CEO of AI PDF... one of the biggest AI PDF readers in the world." ## Intended Audience (as demonstrated) - Dan explicitly addresses an audience interested in practical AI adoption and startup building; he describes his own viewers/listeners as comparable to the guest's users: "our users, because they are like you." - Both speakers characterize the relevant audience as **early adopters and "information nerds"**: "So, you're kind of like one of your core users... it's like information nerds... that's a certain kind of nerd that like cares about that." - Dan also mentions an adjacent corporate audience he serves through Every ("we do a lot of work with big companies where we help them figure out what to use") and distinguishes them from the show's early-adopter framing. ## Observable Presentation Patterns - **Interview-driven structure**: Dan opens with a fact-dense introduction of the guest ("about 500,000 registered users... over 2 million conversations in the GPT store... almost 3,000 paying subscribers"), then probes with follow-up questions ("I want to push you a little more"). - **Reciprocal exchange of builder experiences**: Dan shares his own company's practices (Spiral, fundraising history: "we raised like 700k... and 150k"), positioning the show as peer conversation ("We can swap stories"). - **Live comparative testing**: The episode includes an on-air comparison against Perplexity with the same prompt: "let's do a quick [compar] here... I'm going to get the exact same prompt to be fair." - **Standard creator-education outro**: The video ends with an (apparently ironic/AI-generated) call-to-action to "smash that like button and subscribe," describing the show as delivering "knowledge bombs about ChatGPT" — observable evidence of a YouTube-native presentation. ## Distinguishing Claims from Demonstration - **Creator claims about the show**: mission of spreading AI literacy ("AI is actually for everyone" is the *guest's* claim; "helps spread the word" is the guest's characterization of the show) and the outro's self-description as "the epitome of awesomeness." - **Demonstrated**: long-form founder interviews, live demos, cross-tool comparisons, host's own startup-building commentary, and a named episode-numbered series ("Ep. 39") indicating a recurring episodic format. - No metadata was supplied beyond the title; title confirms the episodic numbering and guest-focused naming convention. # Delivery Observations: "His GPT Wrapper Has Half a Million Users—And Keeps Growing" ## Format Two-person interview podcast (host **Dan Shipper**, guest **Vicente Silveira**, CEO of AI PDF / AI Drive), including **live screen-share product demos** in the middle and end of the episode, plus a pre-produced cold open teaser before the intro. --- ## Tone & Register - **Conversational and colloquial throughout.** Both speakers use casual filler heavily: Dan says "like" constantly ("it's like a it's a different market"), and Vicente uses "you know," "right?," and "kind of" as recurring discourse markers: > "It's kind of interesting. Um, yeah, at some point it was when uh OpenAI for the first time they allowed to upload PDFs..." - **Mutual enthusiasm and affirmation.** Both speakers frequently validate each other's points: - Dan: "That's really interesting," "This is cool," "I think you may have converted me." - Vicente: "Yeah, I think that's totally true," "You're so right." - **Self-deprecating and blunt humor from the host**, e.g., "I guess it's eliminating any chance that Perplexity ever sponsors this show, but um—No, no, no, no. They they do great work." --- ## Recurring Phrases & Verbal Habits - **Vicente's repeated "right?" tag** at the end of explanatory clauses — appears dozens of times ("which is great, but it's actually not the smartest at planning... right?"). - **Dan's "That's interesting" / "That's really interesting"** as a transition device between guest answers and follow-up questions — occurs at least four times. - **Both speakers repeat words for emphasis in real time:** "we we died," "it took off took off immediately," "a a a shot at anything," "the the the technology." - **Vicente's "So" as a reset/restart marker** when reformulating explanations: "So, what do I mean by that?," "So think about this." --- ## Interviewing Habits (Host) - **Stacked statistics in the introduction** to establish credibility quickly: > "You have about 500,000 registered users... You've done over 2 million conversations in the GPT store and you just started monetizing and you have about almost 3,000 paying subscribers." - **Explicit pushback and probing follow-ups:** > "That's interesting, but I want to I want to push you a little more... Do you have a a theory about like why they won't eventually do like a multi-folder upload?" - **Restates and mirrors the guest's points** before extending them ("And I think like the the thing that comes to mind for me that I think you're saying, which I think is true for us as well, is..."). - **Personal anecdotes as parallel evidence** — Spiral, Every's fundraising amounts, fitness tracker app, second-language writers on his team. --- ## Teaching / Explaining Habits (Guest) - **Analogies as a primary explanatory tool**, repeatedly: - The "rapper death countdown" clock for competitor obituaries. - Loom vs. YouTube/Vimeo for why incumbents don't build everything. - Excel as a metaphor for general-purpose AI tools (Dan's version). - The **wealthy person with human agents** analogy, used twice: > "you just think like how a a very wealthy person operates, right? They They operate through intelligent agents..." - **Narrating live demos step-by-step**, pairing actions with explanations and disclaimers: > "This is just kind of a and this is a live demo here. So, anything can happen." He also explains agent behavior as it unfolds: "So, he went for fetching URL again. Oh, he found it." - **Numbered/structural framing of concepts**, e.g., breaking the persona into components: "So one is like they're not just an early adopter right?... The other component of our user base is they have an actual job." --- ## Structural Communication Patterns - **Long multi-clause run-on answers** from the guest, often self-correcting mid-sentence. - **Question → analogy answer → host elaboration loop:** Vicente answers with a story or analogy; Dan extends it with his own example, then asks the next question. This pattern repeats across the funding, persona, and agent-management topics. - **Comparative benchmarking habit:** both speakers repeatedly explain by contrast — ChatGPT vs. their product, Perplexity vs. AI Drive (run side-by-side with the same prompt), big companies vs. startups. --- ## Isolated Moments (Not Established Patterns) - The **outro** shifts abruptly to an exaggerated promotional voice ("smash that like button... Dan, I'm absolutely hopelessly in love with you") — clearly a stylized segment, not representative of the interview delivery. - Vicente's **Brazil/smuggled computer anecdote** is a one-off personal history example. - The **censored profanities** ("they're [ __ ]") appear in Dan's speech, including "big companies like always find a way to [ __ ] things up" — occasional casual profanity, but frequency can't be confirmed as a stable habit from this transcript alone. --- ## Uncertainty Notes - Speaker attribution for a few interjections is ambiguous in the transcript (e.g., around the onboarding comparison, "When you sign up for our product, our onboarding sucks. Your onboarding is beautiful"), so habitual compliment-exchange patterns are only weakly supported. - Timestamps are transcript line indices, not wall-clock times; pacing observations (pauses, laughter) cannot be verified from text alone. # Position & Reasoning Record: Ep. 39 with Vicente Silveira ## Speakers - **Dan Shipper** — host, runs Every (media company) - **Vicente Silveira** — CEO of AI PDF / AI Drive, ex-Meta ## Business Facts (as stated by Dan, confirmed by Vicente) - AI PDF: ~**500,000 registered users**, live since end of last year - **2M+ conversations** in the GPT store; ~**3,000 paying subscribers** since monetizing - Team of **five**: Vicente, co-founder Kartik, three engineers, plus front-end contractors - Raised only a **friends & family round**; runs at **break-even** ## Core Positions ### 1. "GPT wrapper" businesses are repeatedly pronounced dead but can thrive - Vicente: they joke about a "wrapper death countdown"; competitors gave up when ChatGPT added native PDF upload, but his business kept growing. - Reasoning: "everyone is pointing at each other saying, 'You are a wrapper'" — the label is uninformative. - **Loom analogy** (Vicente): YouTube/Vimeo had the technology for async video recording but didn't do it; Loom "nailed the use case" and sold to Atlassian. Similarly, AI Drive nails **end-to-end document workflows over a user's whole file collection**. ### 2. Why users don't just use ChatGPT - Vicente's claims: - ChatGPT/Anthropic/OpenAI are "racing toward AGI" and serve **minimum core use cases**, not specific deep workflows. - Users with large collections need capabilities platforms don't offer: accounts with **150,000+ files**, multi-level folders, agentic file manipulation. - Model-agnostic hub: users want to switch to whichever model (Claude, o1, Gemini) is best; uploading files separately to each is friction. - "We're giving a cloud drive to an AI agent" — very different from "allowing an AI agent to access some files." Chat history itself is files the agent can search and manipulate. - **Qualification preserved**: Vicente explicitly acknowledged platforms *could* compete ("Yes, but that is always the question for startups"). ### 3. Persona definition - Not just early adopters: Vicente defines his user as an **early adopter + risk-taker WITH an actual job to be done involving lots of documents**. Examples: 80-year-old law firm partner, researchers, accountants, writers. - Dan contrasts: big-company clients he works with haven't heard of Claude — "a different market." ### 4. Big companies can't take risks; that's the startup opening - Dan: big companies "find a way to [mess] things up" via innovator's dilemma; must serve the lowest common denominator (fitness app AI example — "milktoast"). - Vicente: incumbents' customers are both their strength and their vulnerability; radically AI-first rewrites of existing products are "too radical" for them. - Demonstrated live: **Perplexity declined to download external content** ("I cannot directly download content from external websites") while AI Drive's agent fetched, parsed, and created documents. Vicente also notes Perplexity's focus is replacing Google as an answer engine — a different focus than document manipulation. ### 5. Lean methodology: ship fast, serve early adopters, evolve - Origin: prompt-injection experiments post-Meta; emailed Greg Brockman at OpenAI; sought a job, asked for an API key instead; started as a ChatGPT plugin running on Replit ("the least effort thing"). - Key pivot moment: they added file upload expecting no one to use it (Google/Dropbox rate-limiting); their domain became the #1 link source within a week — teaching that their users were risk-takers. - Then built standalone AI Drive because users "needed an actual environment where they could do the work end-to-end" and to verify sources. - Dan's stated **trade-off** of this approach: the product evolves so far from the start that marketing/public identity constantly lags reality. Vicente agrees they're "flipping" the marketing now. - Vicente's defense: the ground is shifting (phase shifts like GPT-4 destroying specialized models — his Watson PDF-extraction anecdote; Microsoft PII model vs GPT-4 paper). Early adopters **tolerate experimentation** and pull the product toward the next thing (e.g., agents), letting the company track the market rather than get stuck like ICQ. ### 6. Fundraising restraint - Vicente: fundraising process dragged ("tweaking PowerPoints" like Meta again) while users demanded features; product was then OpenAI-dependent; AI investment hangover. Chose monetization instead. - May raise later, but cautiously — citing **Bill Gurley**: money makes companies complacent; startups will need fewer people and less software than in the past 5–10 years. - Dan's parallel position: Every raised < $1M total (~$700k in 2020, ~$150k recently); AI itself provides the "pull forward" effect capital used to buy; using AI well is rarer than getting capital for technical teams. ### 7. Humans as agent managers ("allocation economy") - Vicente: new hires will manage agents with specific jobs — e.g., a person responsible for an AI onboarding agent (their onboarding "sucks"; the agent will greet users contextually, e.g., lawyer landing-page visitors). He has "two mentors: AI and my co-founder." - Dan: frames this as the **"allocation economy"** — manager skills become more widespread; managing AI mirrors managing people (how much to delegate vs. micromanage; "have you ever managed a person who is literally a general intelligence? It's hard"). - Vicente's counter to "agents do everything so nothing to do": wealthy people can hire people and still mostly fail — leadership, vision, grit still required. Same will apply to agents. - Vicente's productivity example: at Meta, a PM's support-report request took a person a day; AI does it now. ### 8. AI democratization - Vicente: mainstream media messaging (AI kills you / takes jobs / rich get richer) demotivates and strips agency; reality is the opposite — anyone with a phone practices management skills via prompting. Growing up in Brazil, he had to smuggle computer parts; AI is the first revolution where "most people are already equipped" (low bandwidth works well). - Dan: AI immediately upskilled internal staff (fluent English writing for second-language speakers); fortunate that AI is cheap and public rather than restricted to IBM/DOD-style access; "raising the floor creates economic opportunities." - Vicente's aside: "or you're in Europe, right? They're still struggling." ### 9. Agent demo (live, with acknowledged risk) - Delegation architecture: main agent (GPT-4o) delegates planning to an "expert" model (o1 or latest Claude); then sub-delegates file-reading to work around context-window limits. - Agent worked around a failed tool call by retrying URL fetching — "the more tools you give, if there's overlap, they can find workarounds." - Vicente's design thinking: agent tools must be **self-explanatory and do what's expected**. - Dan's reaction: "you may have converted me"; better than NotebookLM because fetching is automatic. ## Unresolved Tensions (preserved) - **Platform risk**: Vicente admits platforms *could* replicate his features; his defense is strategic reasoning (focus, risk tolerance), not impossibility. - **Marketing lag**: Dan identifies and Vicente accepts an ongoing gap between public positioning and product reality. - **Agent reliability**: Vicente concedes agents "still don't work great" — sometimes amazing, sometimes stuck. - Dan notes he's "probably anti-VC biased" — personal framing, not settled fact. # Speaker Attribution & Provenance Record **Video:** His GPT Wrapper Has Half a Million Users—And Keeps Growing - Ep. 39 with Vicente Silveira **Video ID:** 01a0f15b-0595-7b19-8695-00ee7fb6ba34 ## Identified Speakers & Roles - **Dan Shipper** — Host / interviewer - Identified via greeting exchange ("Sante, welcome to the show. Hey Dan") and outro ("with Dan Shipper as the captain of the spaceship") - Describes himself as running a media company (**Every**), employing editors, writers, designers - References internal Every products: **Spiral** and the **Chain of Thought** column (every.to/chain-of-thought), **Context Window** newsletter - **Vicente Silveira** — Guest - CEO of **AI PDF** / product called **AI Drive** (website: myidrive.com) - Former **Meta** product manager; co-founder **Kartik** (also rendered "Karthik"), described as former Google AI engineer - X handle: **@vicentes** ## Additional Voice (Unattributed) - **End-of-video promotional segment** — an unidentified speaker delivers the outro after the guest sign-off: > "Oh my gosh, folks, you absolutely positively have to smash that like button and subscribe to AI and I... Dan, I'm absolutely hopelessly in love with you." - **Flag: promotional / call-to-action segment**, speaker role not identified in transcript or metadata. Likely an AI-generated voice parody, but this is not stated in the source. ## Referenced Third Parties & Claims - **Greg Brockman** (OpenAI) — guest says he emailed him about prompt injection findings and received a reply directing him onward - **OpenAI / ChatGPT, Anthropic / Claude, Google Gemini, Perplexity, Notion, Loom/Atlassian, Meta, IBM, DOD, YouTube, Vimeo, Google Drive, Dropbox, Replit, X (Yohe / baby AGI from a prior episode)** — all referenced as third parties or competitors; no first-party statements from them - Cited sources mentioned by speakers (unverified by them beyond attribution): - A **Microsoft paper** comparing a PII detection model to GPT-4 - **Bill Gurley** quoted on fundraising causing complacency - A company that built PDF extraction on **Watson** ## Demo Material Shown (Guest-Operated) - Live demos of AI Drive agent workflows, including fetching Every articles, expert-model planning delegation, and a comparison run in **Perplexity Pro** (which declined to download external content) ## Provenance & Availability Notes - **Comment availability:** not specified in source material - **Curated metadata:** none supplied beyond title and transcript - **Interviewer's attribution note:** guest's company name spoken variously as "AI PDF," with product named "AI Drive"; co-founder name appears as both "Kartik" and "Karthik" in transcript — spelling uncertain We're nailing the use case of getting a collection of documents that you have and being able to do end-to-end workflows with those documents. Can you go to this URL and download the articles and then create a document with the main ideas in those articles and express those main ideas as sentences. Write the text of the site into the files and then wow, interesting. Okay, so now we're getting some outputs. Sante, welcome to the show. Hey Dan, thank you for having me here. Thanks for coming on. So, for people who don't know, you are the CEO of AI PDF. It's one of the biggest AI PDF readers in the world. You have about 500,000 registered users and it's only been live since the end of last year. You've done over 2 million conversations in the GPT store and you just started monetizing and you have about almost 3,000 paying subscribers. So, it's a really, really cool business and and what I think is most interesting is there's this narrative especially like in the sort of earlier days of AI but I I think it's still happening now which is like oh like yeah, all these like AI PDF companies are like they're \[ \_\_ \] and they're not going to do well and whatever. Um and I think you're actually building like a really interesting business and you're also part of this like broader wave of people who are making things in the in this new AI economy with like really small lean teams. You run the company break even without raising a ton of money. I know you raised a friends and family around but like you haven't raised a ton of VC. I think there's a lot of overlaps between the kinds of things that you're building and the kinds of things we're building internally at Every and I just want to like um learn about how you're thinking about it. We can swap stories. We've done a lot of thinking too about fundraising and and all that kind of stuff and how to how to build this cuz I think people assume oh, you can't build a good business like this but I think you can actually build like a sneaky really, really great business this way. Uh small teams especially with AI, so um yeah, tell us about tell us about that. Tell us about your business and and how you're thinking about running it. Yeah, it's funny you mentioned the the whole rapper thing, right? We we almost have like a like a a rapper death countdown, you know, clock, right? Which is like this many days since we we died the last time that that we keep track of because from from time to time we're pronounced dead, right? Uh in reality, the business uh keeps growing. It's kind of interesting. Um yeah, at some point it was when uh OpenAI for the first time they allowed to upload PDFs to ChatGPT, right? It was like, "Oh, all this stuff is dead, right?" You know, including some of our competitors actually just gave up during that time. Uh but we kept plugging at it and um and I think it at some level, you know, this is an interesting industry because everyone is pointing at each other saying, "You are a rapper," right? It's like, you know, I don't know if Nvidia is a wrapper around math and everything goes goes from there. We started on uh on this because we tried different things actually when uh we first started uh when ChatGPT just was coming out and uh I was basically trying to do like property injection against uh ChatGPT and and Sydney back then. And I found some stuff, I sent it to uh an email to Greg Brockman at OpenAI, and he replied. He was like, "Oh, that's kind of interesting. You should talk to this guy here," right? And at the time I was actually still kind of like I just, you know, got out of Meta and I was angling for a job at uh at OpenAI, right? I talked to this guy. He was like, "Well, you know, you you have an interesting background. Uh you know, right now we're not hiring for this. Maybe in a couple of months, I'll keep you in mind." And I'm like, "Well, maybe you can give me like an API key or let me into this new developer program." And that's how we started, right? And and we tried different ideas, but the PDF one really took off took off immediately. And in reality, it's because uh this is one of the first things that are people are trying to figure out with AI. It's a lot of pain around dealing with lots of documents and PDF is the the main kind of document across platforms. So, people just just gravitated to that. That's really interesting. And so, people so people gravitated to that and and you you know, you've said like people have declared your death multiple times. I mean, if if a PDF reader is in ChatGPT, like why are people using you? Yeah. Yeah, great question. So, the thing is um when we looked at this and we started building it, um we actually at first we didn't even have a place for people to upload PDFs, right? We just like, "Okay, you know, you can uh just give us a link and we'll our server would go go there and fetch the content." And because we're growing so fast at the time, uh people were giving us like Google Drive links and Dropbox links and Google Google and Dropbox started uh rate limiting our IPs, right? Because they saw us like as an aggressive bot. And and and this was a bad experience for our users cuz they would get an error. And we were like, "Well, you know, what are we going to do?" And we're like, "Well, I guess we could try to play the cat and mouse game with those guys." Uh but we know we actually know how that works and that would be kind of pretty bad. Um so, we're like, "Well, what if we just let them upload their their files?" And we thought no one would do it, right? And so, the first version of our website, my co-founder Kartik, he was like, "It looks scary." And to our surprise, like in a week that domain for our website became the number one domain for the links, you know, passing the Google domain and the Dropbox domain. So, that was a lesson for us because it told us that the the users that were gravitating to ChatGPT and they're going to all the trouble to enable plugins, that those people were actually risk-takers and early adopters, right? So, we built that for them. And then kind of going back to your questions like, "Why do they kept using us uh even after this was created uh on on ChatGPT proper?" It's because they don't want to just upload one file. When it works, they basically want to upload their whole collection of files. And even to this date, I think on ChatGPT you're limited to like 20 files, right? And for us, it's like we have people with more than 150,000 files in one account. We have people with multi-level folders, right? No one else supports that kind of stuff, but it's important because people need that low friction and even respecting the intelligence they put in creating their folder structure uh, to be part of of this onboarding, right? So, this part of the the product experience. That's interesting, but I want to I want to push you a little more. So, like it sounds like, you know, uh, one of the reasons people are still using you is because you're you're sort of staying one step ahead of what ChatGPT will do. Do you have a a theory about like why they won't eventually do like a multi-folder upload? Like an example might be uh, Milk Road where it's like it it does let you it does let you open a lot of files. Like are you worried about that? Or is there some strategic reason why why you think, um, you're going to go deeper than a ChatGPT or like a Gemini or or whatever? Yeah. I mean, this is this is interesting, right? Because I feel like that these guys, uh, especially if you look at uh, ChatGPT, uh, their their focus is to kind of race towards AGI and create a product that's uh, good enough to have enough usability there, uh, so that lots of people use they can collect the training data and then feed their machine, right? I don't think they're going into one particular specific direction. They're mostly kind of like touching on kind of like what is the minimum for core use cases, right? Um, so I think, you know, is it is there a possibility that something like, uh, ChatGPT or Claude or or something like that can actually compete with us and and and basically there's there's no need for this kind of platform? Yes, but that is always the question for startups when things start, right? So, when you think about a startup like Loom, what do What does Loom exist and was sold I think almost like a billion dollars to uh I think uh Atlassian. Uh Loom is just recording video, right? But they did that use case so well that, you know, even though YouTube had all the technology to do it, they didn't do it. Vimeo had technology to do it, they didn't do it. All of the major providers, they had the the the technology, they didn't do it. But Loom actually nailed the use case. So, for us, we're nailing the use case of getting a collection of of documents that you have and being able to do end-to-end uh workflows with those documents. Mhm. That's That's really interesting. I'm I mean, I think it dovetails with some of the things that that I think about. Like one of the things that I have in my mind when we build things internally at Every is that ChatGPT and Claude, I think like you said rightly, they're building for um the most broad use cases possible. Um like they just want anyone to go on and be able to do whatever they want, basically. And it's sort of like In that way, it's a little bit like Excel where um anyone can go to Excel and you have a blank, you know, blank new uh sheet full of lots of cells and you can just like start typing numbers and anyone can do that. Um and then people are going to discover as they're using ChatGPT and Claude that um uh that they have specific more specific like they they're going to discover use cases for themselves they didn't know exist existed. Um so, for example, we have a we have a product called Spiral that lets you automate a lot of creative work. Like it helps you do headlines and come up with tweets and and all this all this kind of stuff. And and that's the thing you can do with Claude. Um but Claude is not purpose-built for it. That's right. So, our kind of like thesis is um people will discover use cases um for AI and discover problems to solve with AI by using these more general-purpose tools. And then that will create demand um for other players that sort of peel off some of those use cases for particular kinds of people for particular kinds of workflows. Like for us it's like marketers and creators who have like a very specific need for that kind of workflow. Um and and having a product that's purpose built is going to serve those people better. Um and I'm I'm sort of curious for you cuz a key part of my thesis there is like you need to have a particular persona in order to like be powerful enough for that kind of workflow. But it sounds like you're you're kind of going a a level up which is like more general. Um but do you have a particular persona in mind or like how do you think about it? Yeah, that's and I we get this question quite a bit and it's uh it is interesting because our our persona is an early adopter of technology, a risk-taker that has an actual job to be done involving lots of documents. Right? So so that's like all of those things are important, right? So one is like they're not just an early adopter right? Because an early adopter and a lot of actually ChatGPT and Claude use is people that sign up to just see what's possible. They don't have a an actual job they're going to do there, but they they just want to get that, you know, familiar with technology, be able to talk to someone about it. Those kinds of things. So a lot of a lot of that is is that kind of use case. So that's, you know, one component of our user base. The other component of our user base is they have an actual job that they need to do, right? Today. And they have a lot of documents, right? So this is the combination of things that creates our So like who are these people, right? So we have like a There's a law firm, you know, the the partner is an 80-year-old lawyer. Right? And he found us he's like I'm using this every day and I'm going to have it within our firm and I'm a decision-maker. I'm we're going to have it you know adopted here. And we have people that are like uh researchers, we have accountants, uh we have writers. So there's an a range of these different types of profiles, but what you find in common is that they bring a lot of documents uh to the platform and they are basically trying to get some job done today and you know they want to do this in the new way which is the AI first way. So So that that's one thing on the persona. The other thing that I want to kind of uh highlight is we also differentiate from a platform like uh like Claude or or ChatGPT in a way that I think they're not really going there which is we're building this from the perspective of it is kind of like we're giving a cloud drive to an AI agent. Right? That's very different than you allow an AI agent to access some files. So what do I mean by that? And we can show you a little bit about that as well, but like the the agent that we have in AI Drive is capable of doing things like creating new files, updating metadata in files, kind of going through the file structure, right? So it's really kind of driving that uh effectively that that cloud drive uh to be able to accomplish a job for the user. Right? Because a lot of the job involves manipulating a lot of documents, right? So that's like another very important point which I think it's not really the focus for these other platforms. Why do you think it's not the focus? Because I think it's not it's not necessary to accomplish what they're trying to do. And it also introduces other types of considerations and risks that they may not be interested in dealing with, right? So like this opens up our platform for for a tinker type user, right? To be able to do things like So think about this. So you have a you know chat history in uh something like ChatGPT, right? But that's something that, you know, you can go there, you can look at the chat history. Uh in our product, chat history is actually made of files that the the I can access those files uh on your behalf and use the same tools like a search tool to be able to go into those kinds of files, right? So these are all kind of like the same primitives that we're building on. So you're kind of like one of your core users. You said early adopters, but it's like I mean, I'm thinking of myself like information nerds. I was like, "Ooh, your chats are files? Like that's a like that's so cool." It's like feeding back into itself, but like that's a certain kind of nerd that like cares about that. just going to illustrate what you're saying. So you see like we have this little system folder here. So basically, you're sharing AI PDF and it's it's called the product is called in the actual product itself, it's called AI Drive. And basically, you're showing me on the left, you have basically it looks like you have a list of files, so it's like that's your drive. It's like a Google Drive. And then you have a a chat window, um which I assume allows you to sort of chat with those files sort of like a Notebook LM. And then on the right, it looks like there's a reader view where you can for any particular file that you're talking to, you can also like see the PDF. Is that that's basically what I'm looking at? Yes, that's right. And you can see like on the left side here, we have these kind of short shortcuts here. So if I click on this little uh what looks like a clock, right? And you can see a history, right? So this is very similar that you have in chat GPT. There's there's a history of the chats that you had today, right? And that's all you get in a typical platform. That's not really open for people that are like tech enthusiasts like like you and and and many others. But what what that means for us is that we're building this so like if I go back to files, you can go and find these as files under the chat history folder, right? So we're building these as basically kind of like an open platform where everything from your chats uh to, you know, and we're going to be adding memory, uh all of that will be basically just a file in the system. That's interesting. Um, I want to go back to what we were talking about earlier, like, so one persona is like information nerds like me, and maybe those that like cuts across lots of different industries. Like there's probably information nerds who are lawyers or accountants. Um, it doesn't seem like any of your marketing is like specifically about that. Like your marketing is a lot more like general AI PDF stuff. How do you feel about that? Yeah, and we're flipping that because our our our trajectory is going through this transformation, right? So if you go back, when we built this, and I think this is actually a good, uh, I I give this advice to people when they're when they're thinking about what they're going to do. It's like, what we did with the plugin, that was the least effort thing that we could have done at the time, right? It's just like an API. At the time I had a server running on Replit, right? Replit's amazing. Uh, and that was everything. And we with that we discovered the market, right? But then at that point we're just a plugin. So we couldn't operate independently, right? What do you mean by it was just a plugin? There was no web app, no place for you to create an account. We had no direct relationship with the user. So you would go to ChatGPT, you would, you know, enable plugins, you would find us. I see. It was a ChatGPT plugin. Okay, I just forgot about that whole that whole era of It wasn't that long ago, but it feels so feels like that. Um, yeah, so like so so at that point, right? We we were a product that only made sense as an add-on to ChatGPT, right? Uh, but what we realized is is what? Is that people needed an actual environment where they could do the work end-to-end, right? With the AI and the files, be able to verify the work of the files, uh, which is another thing that we, you know, as you can see on the right side of our, you know, AI Drive screen, you have the actual files that, you know, are the source material. Um, but beyond that, the other thing that we realized and I think it will continue to be true for the foreseeable future is because of this arms race between the the main providers, right? So, you can you can the ones that can play the game right now probably, you know, Google and Tropic and OpenAI, X I guess is coming up as well as a potential one. So, you have maybe like four four five providers there that can actually provide unique capabilities when it comes to the models. And our users, because they are like you, right? And and and you know, I I I heard you say in your podcast you tried this on Claude or this on 01, right? They want to be able to have the latest and greatest. Now, if you're going to upload your files to ChatGPT and then tomorrow Claude is has a better reasoning model, then now you get to upload your files to over there as well. So, we bring that one place where where you can use all the models combined. So, that's the other aspect of that. That makes a lot of sense. I mean, it's funny like we do a lot of work with big companies where we're help them figure out what to use and and and and sometimes train their train their employees and that kind of thing and I think mostly they're not in this category. Like they just they're excited to see that ChatGPT added a new feature, but none of them have heard of Claude. They're like, "What?" Or like it's the best like it's whatever, you know? And um, so yeah, I think there is there is that sort of separate a different it's a it's a different market. The early adopter market is a bit different or for people who like really want to use the latest and greatest type of thing. And I assume that like I said earlier like that sort of cuts across industries and and there's there is room for that kind of like nerd and what's nerd as a customer and and what what's interesting to me about how you're doing this is um, you're kind of like you're just taking the first step first. You're like, "Okay, cool. I'm going to make a plugin." Um, and then that starts working and then you're like, "Okay, like now our customers really need like a place that's like different from the ChatGPT form factor. So, so to go build this other thing and then now you're adding more features onto it and and and that kind of thing rather than I think there's this other approach which is instead of getting to market like super quick you kind of um come up with an idea you make a deck you raise money you start to recruit a team and you take like a year to like put this thing that's like your vision into the world you know and that's a that's an alternative thing I think that works too and there are always trade-offs like a trade-off that I'm seeing here in the way that you built this is the product that you eventually ended up building is sufficiently different from where you started that there's like always this like difference between what is public like what the public marketing is about and how you talk about it and like where the where the product is now and you need to like constantly sort of like catch up I've I've sort of I've experienced that a lot before so that's one trade-off where someone who just like just put the vision out into the world like singularly like that's not their problem their problem is like does anyone even want this um I'm just yeah I'm curious how you how you came to that or like why that's your that's your methodology Yeah I'm glad you asked this because um I you are right you know there's different ways and I have a lot of friends founders and they they took these other paths right um I think one is you know what what works for you right and for us this whole thing started as a side project you know with with me and Karthik and we're just like hacking over the weekend because we just love the the stuff right that's one but um but the other thing that I think about it now and I think it's still a reality for us is that we are even though it may feel like oh you know AI is kind of already sort of happened right and we are so early in this AI cycle in in the in in ground and you and sure you feel that way as well then it's like the the end the steady state of this technology or the productivity state of this technology it's a still not very clear what that's going to look like right so just just in this short time that we we've been doing this stuff you start with like okay you have these first you had these these like specialized models right that people had a bunch of companies and we talked to I just recently talked to a a company where they had a model for PDF extraction that they they built on Watson and all of that and then they tried it with you know GPT-4 and it just blew that thing that they worked on for years out of the water right so so that that was like a big phase shift there where a lot of people had like a rude awakening there there's a great paper by Microsoft when they they did a big off between the Microsoft PII the private you know information detection model that was built over years with Microsoft resources and they baked that against GPT-4 and it destroyed it right so so that so that was the first the big first phase shift but the reality is that that continues to happen right so we we build this thing and you know at first chat barely worked now chat works pretty well right and we have multimodal chat and all those kinds of things and now you know the if you look at what you know Sam Altman and a bunch of people building now are saying the building that the foundation they're pointing to like oh the AIs are going to become more and more capable so they can take more of the task and become these agents right where everyone is talking about this kind of okay we're going to move to agents so the ground is shifting as we go along right we have something like computer use now it's so Claude did a little right a little demo of that. So, what I mean by that is by building this for an early adopter crowd with an actual problem to solve, right? They have day jobs. That's very important, right? Um we are actually capable of tracking the evolution of this market, right? So, that we don't get stuck on a kind of like a early internet type type thing. Like, oh, there was ICQ, there was, you know, a bunch of different things that eventually became uh irrelevant once you got into the productivity state. So, so in a way it's like for us doing this is a little bit also is both strategic and also a defensive as well. Mhm. Well, let me unpack a little bit of what you said cuz the question I'd ask is sort of like that tinker mentality where you're go you're going out and you're just building like the thing and you're getting to market super quick. And it sounds like what you're saying is that by serving an early adopter market, you'll be able to and you're incentivized to kind of like keep up with the latest and greatest so that you don't get left behind. How do you bridge the two How do you bridge the gap between the two? Like, what's the bridge from how you got to market to not being left behind because you're serving early adopters? Yeah, I think I I think the early adopter what they give to you is some leeway to experiment more, right? So, for example, we now introduce the agent uh into our product. Before, if you look at our product, right, and it's still there. So, that's the main way. You can see this menu, right? And then basically, I clicked on this menu in the chat prompt and you have the models from the main families here on Tropic Open AI and uh and Google Gemini. So, you can just go into this and just do a regular chat, which most users that's what they are uh they're used to. But then we have our kind of like the users are pushing us forward. They want to go into the agent and be able to do more of that task, right? So, we're trying to basically work with the uh core of our early adopter users and also listen to the ones that are already moving forward to the next thing. And what these early adopter users, what they give you is more tolerance for the experimentation because of course and something like an agent today still doesn't work great, right? And you may have a moment where it's like amazing and you and the next one that you do, it may not it may get stuck, right? But they actually want to see where that thing is going. So, that's why these types of users, they help us kind of like bring the both like the main use case and the leading use case uh at the same time forward. I think this is like an important related point to kind of this competitive thing that we've been unpacking together. Like how do you compete if you're like a two-person team or I don't know how big you are, but we're we're eight people. So, like how do you compete where, you know, uh against OpenAI or Google or Notion, whatever. Um and I think like the the thing that comes to mind for me that I that I think you're saying, which I think is true for us as well, is um people forget that when you're a big company, you have to serve a lot of users, it's really hard to take risks and for a while there was this like feeling about AI where it was like, "Well, the AI is going to be smart enough that it's never going to make mistakes. So, like uh big companies are going to be able to do anything that startups can do." And I was always just like, "No, big companies like always find a way to \[ \_\_ \] things up." Um it's it's just like it just it's not because they're not smart. It's just innovators' dilemma stuff. It's basic stuff that you just like can't take risks. And and I think like I think that's why when you look at a lot of the AI stuff, um like for example, and I won't name names cuz I don't I don't like \[ \_\_ \] on people directly, but um for example, I I got a fitness tracker app recently and I it's really great. Like I love actually the way it works and and the app and whatever, but like they have an AI feature in it, and the AI is just like it's so milktoast. Like it just like doesn't say anything useful, basically. It's like and the reason is they have to they have to make it work for like the lowest common denominator user. They have to make it not confusing, and they don't want to take any risks cuz they're like a big company, and it would it would be bad if it said something that was risky. Um which makes a lot of sense, but it means the experiences you're able to do as a bigger company are are less good in a lot of ways than the experiences that you can do as like a as a small company that you can just like decide, "Okay, yeah, we're going to serve these users that like don't care if there are rough edges, and we're going to like explore on the boundaries of what's possible, and our users are going to understand if we return a result that's like not so great, and that allows us to experiment." And they understand cuz they're they're like want the greater power, and they understand that there's a trade-off there, which I think I think that's really important. I think people miss that all the time. Yeah, I think that's totally true. The only reason why startups have a a a shot at anything, it's it's because there's there's some I guess core vulnerability to established business, and usually that is their customers, right? So the the thing that makes them powerful is that they have lots of customers is the same thing that makes it hard for them to take risks with those customers, right? So if you think about a you know, the mainstream products, you pick a spreadsheet product, right? Uh a product like that, there's a lot of investment that was done on a ton of features in training of users. I mean, they actually have, you know, certification courses for their users. Their user base goes there day in and day out expecting a certain experience. They also want to know that there's some AI on the side, and you can put some AI on the side that they do, sprinkle AI everywhere, but to radically change that experience to be something like, "Oh, it's going to be AI first." You're not going to go, you know, click the buttons. You're going to tell an agent to go do for it that for you. And that's going to be the core of the experience. No distracting anything. That is the change that's just too radical, you know, typical for the incumbents to be able to do. So, that's why you have the opportunity for for a you know, a startup like like yours and ours to come in and introduce a new way of doing things, which I think is part of what we're trying to do, which is what is going to be the new way of doing things, right? And we feel like it's a lot more like we have this this analogy here. So, you just think like how a a very wealthy person operates, right? They They operate through intelligent agents, right? People that they hire that are very smart. They learn everything about them. And they operate all the complexity, you know, behind them, right? And so, that's kind of kind of like where we think things are going with AI. Of course, right now it it's it's still far from that. But it will approach and get very close to that. So, this opportunity to build for the new experience, I think is is in leaning into that, right? Is very important. I totally agree. Um I think that that metaphor is so powerful. It's something that I I've written about a lot and I thought about a lot. It's like if you want to know where things are going with agents, like, yeah, people have been hiring agents for a long time. Um and they're they're solving problems that uh with these agents that like there's a lot of overlap with what AI is going to be able to do. And maybe you can do new things with AI that you couldn't do with like hiring a personal assistant or whatever, but like there's a lot there that that you can just like kind of carry over. Like even for me, you know, I I I run a media company and um I employ a lot of people, like, you know, editors and writers that help and uh designers that um help me do things at a high level frequently all the time. You know, this YouTube video is going to be edited and it's going to have, you know, it's going to have an intro sequence and it's going to have a thumbnail and all that kind of stuff. And I can do that because the company is successful enough that I can hire those people. But it took me a long time to get to a scale where I could do that and I think um if you want to understand, for example, where the future of media is going, it's it's not that teams of creators are going away. It's just that an individual creator is going to be able to do a lot of the things I have to hire people to do like on day one. Um and you know, I'll still have lots of people doing stuff. I'll just be doing it at a higher scale cuz the people that are editing my videos can edit twice as many videos or whatever. Yeah, I think that that's I think that's a really good place to look for ideas. It's like if you want to understand how people are going to run their calendars or their emails or whatever, just like look at how people with assistants do it. Um Uh so, yeah, that's definitely a metaphor that that we use a lot internally. I'm curious for you like you raised only a a friends and family round. I assume with the kind of traction you have, you could have gone and raised a venture round. Uh why didn't you do it? Yeah. I think first our experience uh raising was kind of very interesting, right? The beginning was like, "Oh, this is amazing. We're just going to be, you know, we're going to raise a ton of money now and maybe we should raise more." But the beginning was fast and the process then later started dragging along, right? And I felt like I was, you know, back working at Meta doing PowerPoints and tweaking PowerPoints and prepping with a friend, you know, a friendly VC to to talk to another one. And when we had users, they're basically asking us to do stuff. And at the time it was just Kartik and I, right? So, I'm like, I'm hating this, right? And if we if we can just go and monetize, let's just do that, right? And then we'll come back. So, I think that was the that was the main reason and it was a time as well where uh with AI the the whole kind of like hangover of the first wave of investments was what set in, right? So, people were really worried about this. Uh it was a time also for us product-wise that, you know, the product was very much dependent on OpenAI and ChatGPT specifically, right? Which is not the case uh now anymore. So, for those reasons we're like, "Yeah, let's just focus on the product." Uh which I think it was the right uh the right thing to do. That's interesting. Do you think you might raise again in the future? Or like what's what's the path look like for you? Yeah, I think so. I I think the reason is we want to be very careful in in um I guess diligent about when when we do it why we're doing it, right? And and how how we're going to use the the capital that that gets invested. And and this is one of those things where I'm kind of going back to how many how many people and how much software, that's the other question, a startup is going to need moving forward. And we think it's actually less than what has traditionally in the last, you know, 5 to 10 years uh been, right? So, we want to make sure we we do it right so that we we're not like uh Sometimes, you know, I and I and I hear that. I think uh Bill Gurley famously said that is companies raise money and that because they have that money, then they end up becoming more complacent, right? So, that's one thing that for now we want to make sure we don't do. But but yes, I think it was once we know like, "Okay, this is going to allow us to grow in a in this this direction." Uh it will make sense, right? I think the other thing that I want to, you know, maybe touch between this point and the previous one on um how both the experiences different, but also how end up saving money, right? If you just lean in is So, we think about like let's say a job like onboarding on our product. When you sign up for our product, our onboarding sucks. Your onboarding is beautiful. I I did this spiral one. It's beautiful. Thank you. And what No, I have to love it. Yeah, I I love it. Ours sucks, right? And uh we're like, "Okay, we need to make our onboarding better." Um I do think that the way I want to do that is basically given AI a job to do that onboarding of that user, right? So, what does that mean to us is like instead of having sort of a either a product that we attach to our our app that will kind of like be configured for that onboarding in some way or or build that ourselves, we are going to basically lend you an agent, right? That knows about the product, knows where you came from. Oh, you know, we have a landing page for lawyers, right? You came from the landing page for lawyers, so you're likely a lawyer. Um hey, you know, this is what the product does. Uh do you want to, you know, upload one of your files and we can I can show you what it what it can do for you. And it basically give that AI a job. And what we want to do is So, we have a person that we just hired that is going to be responsible for this area. So, that person's job the person the actual human that we hire is to basically manage that little agent, right? And be responsible for the the delivery uh that that little agent does, right? And of course, as the agent gets better, there will be more and more that it will be able to do. So, it's kind of a niche-y way to think about these things where we feel like that as we hire people, they will end up be responsible for certain agents on the product that have specific jobs uh both for the user and for the company as well. I love that. I think that makes a lot of sense. Like I've been writing a lot about you know, what I've been calling the allocation economy and I think this is like right on that on that train which is like in an allocation economy you're instead of doing a lot of the IC work you're you're you're doing a lot of more management work where you're managing the allocation of intelligence, managing agents and um and in in that world like the skills of managers become more important than they are now and they need to be more widely distributed. So I think I think that makes a lot of sense. That's really interesting. I'm curious for you do you have any sense like quantifiable sense of um you said earlier the things you can get done with a smaller team with less capital you can get a lot more done now. Do you have any quantifiable sense of what that is versus like 10 or 15 years ago? Oh for sure and and you can see this everywhere right and I can give an example like working at at Meta, right? Of course Meta has a ton of money, right? And pre gen AI and and now they're deploying it very aggressively internally as I hear from from the outside but pre gen AI you would have things like a product manager and I was a product manager there they would want to know like okay you know what's going on with this particular feature? What are the the top issues that our customers are having in this particular you know area? And so the you would the PM would basically talk to a person in the support area which their job was only to go collate all this feedback and create this report and it would take like maybe a a day or you know if they're busy maybe a little bit more depending on how priority is your problem how much of a priority. So all of that now, right? It's like you can get done with with AI. That's that's just there, right? You can you can get that directly with AI. So, that's just one example of how with the tools that we have now, we should be able to be a lot more efficient and do things that would only be available to to larger companies. Yeah, I think that's true. I mean, we see some of the apps that we all work on internally like I can see someone take something from zero to it's a fully finished product that would have taken a year and they can do it in like two or three months if they're good and they're they have a whole general skill set and we have a bunch of support for them and all that kind of stuff, but it's it is kind of wild and I think I probably have a particular like anti-VC bias or I've been historically hesitant to raise. Uh we raised a little bit in 2020. We raised like 700k and we raised a little bit more recently, but it was like 150k. It's like amounts where it's like a VC would like laugh at it and be like, "What are you going to do with that?" And for me, I'm like, "We've raised less than a million bucks and if you look at the number of products we've built, there's a lot of there's a couple different products and companies that have come out of just that one raise and I think we'll in a year we'll have a bunch more. And I think about raising capital and why you do that, it's you know, if you if you spend a couple months raising, theoretically, uh you can hire more people and spend more on growth so that you can pull forward the progress that you would have made over the next year and maybe make that progress in like three months or or whatever. And I think that that equation it's still there, but it's different. Like it's the amount that you can get done is you basically have gotten a lot of that pull forward effect from just like properly using AI. Uh which which is really interesting and of course everyone else has that. So, um to some degree like having extra capital can help, but capital has always been capital has been pretty available for technical teams and and whatever and I think it's rarer to actually be using AI well than it is to like get capital in certain areas. Like some people have a very difficult time getting capital and that's a whole different problem, but if you're like a technical Silicon Valley-ish team, I think a lot of those people are uh just feel like they don't use AI that much in their daily work cuz they're like, "Oh, I'm better at it." or what I'm better than the AI or whatever. I think that's changing a lot um and people that have like really gone headfirst into it are quite a bit more productive. Yeah. And and uh for us it's like productivity uh at all levels, right? So, I I have a software engineering background. I coded like early in my career and then I stopped. Pretty much went to the business side. I thought I would was never going to be able to code again. And when this came back with GenAI, it was like it was kind of uh you know, you can't mount the bike anymore cuz you can't go up the mountain. Now you have an electric bike and you know, up you go, right? It was incredible and uh and and then it's I say I have like two two mentors. One is, you know, AI the other one's my co-founder uh and I see that even for him, kind of world-class engineer, you know, former Google AI, uh it makes him so much more productive. So, it's like everyone goes up. Like for whatever level you are at, you become a lot more capable. Um so, that's that's absolutely true and we think that there is a huge lack of awareness overall uh in the population. I think some of this is and I like your show because uh it helps spread the word, which is AI is actually for everyone, right? Like people are like if you just follow, I think, the more mainstream media, you think that well, one, AI is going to kill you, but if it doesn't kill you, then it's going to, you know, take your jobs and by the way, the rich will get richer. So, so you end up like what kind of message is that? What does that do to the population, right? Everyone gets demotivated. You take agency out of people. On in reality, you have the opposite. You talked about like this this like learning to be a manager. Well, if you if you have access to an AI, you're practicing being a manager. If you have a phone, you have access to AI. So, you can you can start like specifying a task. The AI doesn't do what you want. You're like, "Hmm. Or actually, my question wasn't good enough." That's a lot of what being a manager is. So, you get your your prompt your question better. Now, you have to look at the result what the AI brought back. Is this quality good? Am I willing to put my name on this thing, right, that the AI came up with? So, so yeah. I think I think that makes a lot of sense. I mean, you know, obviously, there's a lot of difficult social and economic issues to like, you know, broad AI rollouts, but I I do think that point is totally right is people tend to miss how powerful this is as an as an immediate upskill for lots of people. Like, even people we have internally learn. You know, for example, if English is their second language, like maybe they speak fluent English, but their writing of English was like not as good and you could tell and that limits their ability to like get hired or get promoted or do certain jobs. And like the minute ChatGPT came out, it was like a total shift. You know, they could immediately write like fluent English and that level of opportunity like it just opens up that they was was not available before and they didn't have to do anything. Um and I do think that that's that's one of the things I would love to do with this show is show people the easy sort of ways that they can get started and and also like where the what the most interesting or smartest or furthest ahead people are doing so that we can like bring forward like everyone else to use it cuz yeah, I think um hopefully, raising the floor like creates a lot more economic opportunities for people. And and I love that you have this like tutor in your pocket that I would have just talked to ChatGPT all day if I was like 11 or whatever and and I think we're kind of lucky. You know, there's a lot of concern about AI companies like racing ahead and releasing it to the public before it's ready and all that kind of stuff. I think on the other end of the scale we're kind of lucky that we live in a world where all these companies are trying to make it as cheap as possible for everyone to use. Like there's an alternative timeline where it's like IBM invented this and like only like the DOD gets access for the first like 15 years and like that would suck. I don't know, you know, like I I could just I could imagine a lot of things being written the other way where it's like well, only like rich big companies get access to this like crazy intelligence and I'd rather just like have it have everyone If we have to pick I'd rather I'd rather this world, I think. Does does that make I don't know. At this point I'm like I'm totally off off what we usually talk about in this show, but I think it's I think it's really interesting and important. Yeah, I I think you're totally right and or you're in Europe, right? They're still struggling Yeah, I hope that a lot of people get to see this, right? Because it's the first time that I see and I've seen these other, you know, tech revolutions where most people are ready equipped to be able to use it. Right? So, before it's like with, you know, the the microcomputer was super expensive, you know, I I actually grew up in Brazil. We couldn't get it. And and and we we had to basically smuggle parts and and kind of build the Frankenstein computers there. And you know, even with cell phones when they came out, right? It took time until people had access. Then the networks were not good, right? And then now we got this thing that, you know, if you can get whatever social media on your phone, you can get AI AI actually works better cuz, you know, most of it is is lower bandwidth. So, that works well. And I think, you know, for us is is what we see as we go into this kind of like this direction of from just conversations with AIs, right? The AI asks a question, gives me an answer, right? Which is the initial like we chat to, you know, moving to agents, which is I give it a task and it executes, you know, that task for me. Is is this evolution of possibilities that you can that you can have, right? So, like one of the things that we spend a lot of time now is building tools for the agent, right? And it's kind of super interesting because you start to think about like is is my tool good? Right? Is like is is the tool self-explanatory? And does it do what it's expected to do, right? And so, you have that that second that second level there. And and all of this is is opportunity because if you think about like people say, well, you know, when the AIs do everything, then there's nothing to be done, right? But again, if you use the rich, you know, wealthy person metaphor, you can be very rich and you can hire a bunch of people to create a company to do something. But most of the time that doesn't work, right? Because it takes, you know, leadership and talent and some vision and some grit there to be able to organize these human agents, right? Into a into a company that actually succeeds. And I think the same thing will play out with with the agents, right? You're so right. Like I was I'm I'm literally like writing an article about this today, but like it's it's like this there's this weird fallacy where people are like, "Well, agents are going to be doing it, so you won't have to think about it." And I'm like, "Have you ever managed a person who is literally a a general intelligence? Like they are really people are very smart, you know, it's hard. It's really hard and it's it for sure is like a different kind of thing than doing it yourself. But even if someone else is doing it like the skill of delegating like the like going from There's this thing that early managers you have to figure out is like, okay, how much do I delegate and how much do I micromanage? Because if I if I micromanage, it'll get done the way I want it to get done. But if I delegate, but but then I have no leverage and I'm like just basically doing their job for them. So why did I hire them? But if I if I delegate, then I have more time to do other things or think at a higher level, but it comes back wrong. And that's like literally the the problem that a lot of people are having with AI right now is they're like, oh, it sucks or whatever. I have to do I I could just do it quicker myself. And I'm like, that's exactly what managers face. And and so a world where we have these cool let's say it's AGI, like really cool agents that do all this stuff. I still think there's a lot of you're skipping a lot of invisible things that have to be done to like scope the task and pick the right resource and like have the taste and vision to be like, this is what I want done in the world. And I don't I don't think like, okay, maybe we'll get to a point where even they're doing even that, but then there's there's a lot of stuff that is sort of invisible in an AGI type scenario or even before AGI just like intelligent agent scenario where there's a there's a lot of skill and talent that needs to be directing them and I think you're pointing to that. Yeah, I we actually think a lot about delegation. I can show you something cool. I I would love to show you cool because I think that there's a lot of talk about like agents and and and there's a lot of stuff happening, but I don't I haven't seen anything like really compelling yet. Like we had we had Yohe on last week and I think he's got a lot of cool stuff with with with baby AGI, but he's like I think a lot of that is still like pretty prototypey and experimental. So, I'm yeah, I'm really curious to see what you guys are working on. Yeah, let me show you some This is just kind of a and this is a live demo here. So, anything can happen. So, okay, I did download a few of your posts. I think the context window series of articles, right? So, I just want to set the scene for people. So, basically we're back we're back in AI Drive. We've got the same kind of three column layout, which is like we've got a you know, a folder structure on the left, we've got a chat in the middle, and then we've got an open PDF on the right, and you downloaded some Every Articles, some from Context Window, which is our Sunday Digest that goes out every Sunday. Um and you're typing into the chat, and I'll read what you type after you're done typing it. Yeah, so um I'm just here in AI Drive, and I'm just typing this you know, prompt. Uh hi, I'm talking to you Dan Shipper, so writing this to the AI, which is our agent in AI Drive, and he has a series of blogs Context Window in the folder, and then I'm pointing it to which folder I I have downloaded it. Uh can you read them all and suggest some interesting uh talking point or conversation starters for us. Uh and I'm going to use this thing here too as well, which is Use the Expert uh to plan, right? So, this is one delegation too. So, the reason and I'm going to fire this up as I explain. So, now it's kind of processing. And if you click, you can see what it's actually doing. So, first thing it's doing is getting a tool plan from the expert model. So, the main agent you're talking to here is GPT-4o, which is great, but it's actually not the smartest at planning. So, what we do is we allow this agent to delegate to an expert in planning, which can be 01, which is great, or it can be the latest cloud, which is also very good. And you can see this here, right? It says, "Oh, you know, this is the the task. I need to read all the blogs here. Give me a plan." And then you get a plan back. And yeah, so now it's just outputting the conversation starters, right? So, as you can see, let me just collapse here so we can look at all the tools that were used. So, the first thing that it did is get the the plan from the expert, right? Then with the plan, it went about executing it, which is, "Okay, let me take a look in Drive at what's in, you know, in in in the folders and files." So, it it went into the context window folder just to see which files were there. Okay, great. So, now, because and this is one of the things that you have to do with AIs is that each AI has an amount of context window or, you know, working memory that it can work with. And you have to work around that, right? So, here, uh this particular model, well, GPT-4, that family doesn't have that big of a context window. So, in this case here, we basically tell it like, "Okay, you're going to delegate the task of reading these files, right?" So, it went ahead and said, "Okay, go into this folder and extract the key arguments, unique insights, personal anecdotes, blah blah blah, for you and organize this as a conversation starters." So, and then you have the output here, which, you know, it brought back into the chat. So, and then we have here, I don't know, Apple integration to ChatGPT. So, there's like a some interesting conversation starters here about the Apple integration there. So, this is just to show like how in one chat now with a smart agent, you can actually have two delegations that are happening to get that task done. I get it. I think that's interesting. So, which articles have you downloaded? Uh we can we go we can go take a look. So, here and I'm going to drive into the folder for context window. It's the Apple took the stage for WWDC, blue sky thinking, creator led businesses, generate generative thinking, spiraling out of control, and your AI research assistant. Cuz one of the one of the tasks and we can try it on this, I don't know if it'll work, but one of the tasks that I really want to do and I don't think these models are capable of it yet, which is putting everything in the context window, is I want to create like a list of all of the like thesis statements or all of the ideas that were like like, you know, for example, I have this allocation economy idea. That's like an ongoing thread that I write about and I really want to create like a list of all those threads, um, just to like have so I can, you know, refer back to them and this seems to be good thing to do that with. Do you think we could try and see if it it can pull out things like that? Let's give it a shot. So, where do you start from? Where is the information? I would go into chain of thought, which is my specific column on every. Um, it would be every.to/chain-of-thought and download. Let's let's try one thing real quick here. Uh, can you go to every .is.to? .to, yeah. /chain-of-thought. And find that chain of thoughts. Um, and just to make sure this is the website here. I have in chain of thought column URL. And so, we're starting to introduce, uh, yeah, so we're going on the web, right? Uh, so we're using a tool that is a fetch URL content. So, it's going at this every.to, uh, so Dan's, uh, website and let's see if it's going to find the right URL. Well, apparently found something here. Um, oh, it's actually going into AI Drive because we may have downloaded something. So, this is one of the interesting things you see with uh with the tool, right? It's a tool to go out to fetch URLs, but even files here on AI Drive are URLs itself. So, he went there. Let's see what he did. Oh, okay. There's some issue here. Let me try another approach. That's one of the cool things about the agents, right? Is that they kind of work around issues. And the more tools you give, if there's some overlap between the tools, they can actually find workarounds. So, let's see what he did. So, he went for fetching URL again. Oh, he found it. Uh that's awesome. Uh so, so what I wanted to do is uh there's a there's a place there's a way to sort it by newest. Okay. And so, I would like it to just uh sort by newest and you know, take our the last I don't know, all the articles that are on the first page of newest and tell me like, okay, what are what is the what are the main ideas um of each article as as bullet points. Okay, we can give it a quick shot here. So, you want the the the latest articles there? Yeah, so basically there's a there's a it's a new URL. So, I'll just send you the URL just to make it a little bit here. It's just a parameter where you just sort by newest. Got it. Okay. So, basically I want it to go to that URL. I want it to go to each article in the URL and pull out the main idea of the article and express it as a sentence. Um and then give me a bulleted list of the main ideas. Can you go to this URL um which is the chain of thought URL and download the articles um and then create a document with the main ideas in those articles and express those main ideas as sentences using the expert to planner expert planner task. So, it's basically like going to the expert model, which is I guess 01. Yeah, right now it's Claude. And Claude's basically like creating a plan. Uh so, it's writing a plan for the for the other um AI to follow. Um and it's, you know, giving it some, you know, things like it's going to fetch the URL content and write to the file and uh you know, all that kind of stuff. It's telling it, "Okay, here's some things to like watch out for." And now it's starting to take action. So, we're starting to see it, I think, um go to the go to the site um and then write the files into into another file. So, write this write the text of the site into the files. Um and then, wow, interesting. Okay, so now we're getting now we're getting some output. So, the first one is how to figure out what people want is the article title. Understanding customer needs involves thinking in sequences. That's right. Opening AI launches the document code editor. And it the main idea is Canvas enhances human AI collaboration by allowing real-time document code editing. This is cool. Yeah, this is actually quite helpful. I I think you may have converted me. I created that. This is This is definitely different than I think the experience, for example, like one of my problems with Notebook LM, for example, is I have to like go do all this manually, um like uploading all the stuff manually, and it's really nice to uh uh to have it just be able to fetch it for me. And I think the other the other thing that's nice is like, yeah, there's some more complex workflows you can build here that is helpful for text processing. I I like this. This is cool. Yeah. Uh what we find is um it's kind of interesting if you think about the evolution of these things, right? Uh something like if you do a Google search, uh you can't do something like this Google search, cuz usually Google searches are optimized for like maybe two words. I I think that's the uh even if a perplexity You want to try this in perplexity? Sure. Yeah, let's Let's that. Let's do a quick big off here. Um Yeah, I'm going to get the exact same prompt to be fair. Well, I mean, might as well, right? Um So, we can go to Perplexity here and uh oh, I shared this tab. Okay. So, can you see Perplexity here? Yeah. I guess it's eliminating any chance that Perplexity ever sponsors this show, but um No, no, no, no. They they do great work. We we we like them a lot also. Make sure you should do pro. Yeah, make sure it's pro. Yeah, pro. All right, let's fire. Okay. Okay, so it's it's doing some of the um it's, you know, saying, "Okay, we're going to navigate to the URL." It did that. It's going to extract the main ideas from each article, and then it says, "I apologize. I I cannot directly download content from external websites." See, this is exactly what I'm talking about with the risk thing. Is Perplexity is a huge startup now. They're all eyes on them. They have to make sure that it works for everyone and that they don't get sued. And um that means that they have to limit like the edge of the product that they can build is not the edge of the technology. It's the edge of what they legally are allowed to do and like what they think can work for um like the greatest number of people. And because you're, you know, I don't know how many people are you now? We are very small team. So, we have basically it's Cardigan I and another three engineers. We have a few contractors for some front-end type work. five people. And so, what you can do, which is great, and what we what you can do with five people is just so much different, and it's it would not get easier if you had 100. It would get harder. I think that's a really important thing that people miss about this type of product. Yeah. And it's also because I think Perplexity's focus is is really kind of like replacing uh Google, right? As the answer engine versus the question, right? Uh which is which I think is a great angle. Our focus is not that. Our focus is like what you have, which is I have all these, you know, different kinds of documents here and I need to manipulate and create new documents from it. And usually what that means is you have to be really good like you have to have tools that are really good at fetching, parsing, manipulating, creating documents. So it's a kind of like a different focus as well. Yeah, totally. Um well, this is really really cool. I'm I'm really glad we got a chance to chat. Um before we go, if people want to check out what you're working on and just check out your stuff personally, where can they find you on the internet? Yeah, so our website is myidrive.com. That's basically the app that we've been using here during the show. If you want to use if if you have ChatGPT Plus or Teams, you can see our GPT, which is in the productivity category. You can find me on X as well. And I think my handle is @vicentes, so v i c e n t e s. That's my handle on on X. And you know, it's awesome to be able to discuss this kind of stuff with you, Dan. Thank you so much. A true pleasure. Likewise. 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 instead of gold, it's filled with pure unadulterated knowledge bombs about ChatGPT. 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. 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.