Notion’s Founder Deleted 3 Years of Work. Here’s Why.

In 2015, Ivan Zhao and his co-founder moved from San Francisco to Kyoto, Japan, where they spent 18 hours a day in a 2-story house so small that only a traditional shoji screen separated their bedrooms, rebuilding Notion from scratch after scrapping 3 years of code. It's one of tech's great reset stories, but it's also a window into a radically different way of thinking about software.

Today, Ivan runs a company with over 700 AI agents working alongside roughly 1,100 employees at Notion. But this isn't a conversation about AI for AI's sake. It's about what happens when you treat computers not as industrial machines, but as materials to be mastered, like steel, like steam engines, like the fundamental elements that reshape entire civilizations.

Today, we're asking, How do we design organizations, not just tools? How can we think of human scale and centeredness in this age of AI? And what can Renaissance Florence, the design of cities, and Douglas Engelbart's vision of augmented intellect teach us about building in an age of infinite minds? Without further ado, welcome to Possible, Ivan Zhao. Ivan, I've been looking forward to this for a number of months, so welcome to Possible. And we're here at this Unsurprisingly, very cool office. And again, not surprising given the design background, given the focus on kind of the human touch of this. How did you also deliberately curate the office? What's the kind of design and the kind of artistic sensibilities that you bring to the office and to the kind of the culture of the company?

Artistic sensibility. That's it. It's very serious.

Yes. We can use less serious words.

Oh, no, no. This is fine. Artistic sensibility. It's kind of intuitive, I would say. Just like since Notion was small, we also care about what's surrounding us. You don't want to hear noise, but you also don't want to see ugly things. So if we have a choice, we prefer, at least personally, I prefer to see a beautiful Eames chair. And yes, they're a little bit more expensive, but it's extremely comfortable on your body and comfortable on your eyes. That's one factor. The other factor is we're quite inspired by timeless tools in history.

Mm-hmm.

So why not surround yourself with timeless office tools so we can design timeless software?

Yeah, no, it makes sense. So like one of the things I heard about is apparently there was a rug from your childhood that was in Notion's first office.

So it was until like our last office until maybe 6 months ago.

Okay.

I grew up in that rug from my hometown because my hometown was in this Muslim part of China. So we use a lot of rugs. Just there's this red and blue rug we've been ever since, been using ever since the first Notion office and has been traveled with us through all the offices. And 6 months ago we moved to a new office in SF. So I finally took the rug home. Yeah, it's still clean.

That is awesome. So I want to dig right into Notion. And so can you talk about the core problem that Notion AI has tried to solve and has that changed over the last few years?

A lot of people think Notion is a productivity tool, AI-powered workspace. We actually start more from a more philosophical angle with tools and technology. I was really inspired about the computing generation in the '60s and '70s. You know, the hippie generation took acid, thought about, okay, what on the West Coast largely, Yes. what should you do with this mainframe computer in the basement? that's turning out, printing out paper with numbers on it. But if you connect this machine with a monitor display, you can make it interactive and can become a new type of medium. So I was reading their paper in the last year during my college and I realized this is the most meaningful thing I can do by being a programmer, by being a designer, bring this medium of computing to more people. Democratize it, right? So it's no longer just the digital scribe can use that medium, but more people can too. So I've been working on Notion as an idea more than a decade. Started as this kind of you can do everything product. In the last 3 years, AI is this kind of you can do everything new technology. So it's become a new piece, a new Lego block in the toolset for us.

And so in some ways it feels like we are a million years from that time in the '60s and '70s. And other times it feels like we're right back in Xerox PARC. We're creating this new technology. We're so excited. You've said that AI came about at sort of the right moment. Like, how did you know it was the right moment and sort of not too early, not too late? How is it just right?

If you use it, at least the first model that clicked with us, with me, was the GPT-4 class. GPT-3.5 class is like useful, but if you see GPT-4 class model, it's like it's not just a text regurgitator.

Yes.

It's a peace of mind in there, piece of little human thought in this thing. And it's a brand new material that's just like relational database, just like bitmap display. It's a new thing that can just unlock so many new possibilities with it. And there's a funny story that when we got early access from GPT-4, because our friends were working at OpenAI, We thought everybody got access to it as well. So I was like, holy shit, we're going to race to the world to build the first product. We actually launched our first Notion product a week before ChatGPT happened because we're just rushing for it, right? And the story was my co-founder Sam and I got access during a company retreat in Mexico, Cancun, and we just locked ourselves in the hotel room the entire company retreat except the keynote I had to do. Then we're just building the first prototype. And it's, it's a, if you build tools with this, you know, it's a completely different material.

Yeah, no, I remember the GPT-4 moment because I was on the board of OpenAI at the time. And when I got, when I saw the difference in 3.5 and 4, that's part of what made me create the, the, the book before the last one, Impromptu. to try to get to show AI as a collaborator in, in writing a book, that piece of the human mind as ways of doing it. Because as you probably remember, the GPT-4 launch date was kind of moving. So we ultimately ended up publishing it through Amazon directly to be able to, to hit the date, to be on the exact right date. Because part of the thing was I wanted to show what was possible with GPT-4. Now, when you think about Notion has this kind of textual workspace.

Mm-hmm.

But part of what, you know, brings up— and a tool for everything— but part of what comes is one thing is obviously AI doesn't just change the what is a document or what is the kind of equivalent, but also changes interface modalities, right? So like, you know, last year and trying to get people to use AI more, it was kind of like voice pilling. So how do you think about how the nature of kind of the Interface to computing changes. What's the fabric? Yeah, that Notion's becoming and adding that in.

Yeah, rather than consider as a fabric, we like to think in terms of building blocks. Yeah, and to me, it's very difficult for one company to change the building block or language of any anything that humans use. Software UI is kind of just like a language, right? It's like we all grew up using graphic user. interface, we understand there's a box, you're supposed to click on it. That will click mean a different thing. It's almost like speaking French or English. It's a language you learn.

Yeah.

And there's all the building blocks within the language. In some sense, the premise of the first version of Notion was, okay, what are the core languages of computing? Can we— which has historically always been in the hands of programmers— can we open up to allow non-programmers to stitch it together, those building blocks. The building block being text, you mentioned one, text editing, relational database is one of the most powerful building blocks, language, tabular format, different graphic user interface pieces, right? Normal people should be able to deal with that. I would say the constraint, even AI opened up a lot of possibility, but constraint is still there. Like still have to look at a computer screen to interact with whatever thing is. There's more modality introducing through voice and sound, then that's new. But the higher bandwidth thing is still looking at it until we invent something, popularize the brain-computer interface, which is not far probably. Right. So that's another topic. But until then, there's a constraint, which is human biology of seeing things, touching, clicking on things and talking to things. Language model changed that a lot, but still we're constrained by human biology, constrained by our cultures of understanding how to interact with those boxes on the screen. So we sort of can see where I'm coming back to a circle a little bit. So if you think about the first version of the popular AI product is the chatbot. What's the previous version of the killer app before chatbots? Google is a chatbot. It's a text input.

Yes.

That's how we understand this new tech. The most powerful new technology we have, language model. We mirror looking at the image of Google, right? And if you think about in the past couple of years, a year or so, coding agents become really popular, IDE become popular. So everybody was running coding agents. But in the past 6, 9 months, people realized you have a limited bandwidth managing one coding agent. When you're managing a dozen coding agents, what do you do? It's, it's a freaking Kanban board. We're going back to a project management software. So we're, the constraint, it's somehow changed quite slowly because humans don't learn as fast because it's really hard to change the physical display in front of you.

How is the kind of the flow of work at Notion created in terms of humans, agents, which things agents do, which way are amplified? Between them, and where do you think that's evolving to?

Yeah, I don't think anybody knows. I think that's the honest answer. Nobody knows the right answer. Everybody shares practice on Twitter, on blog posts. So we're figuring out this together. I would say the overall trend is if you think you have a company of 10 people or 100 people or 1,000 people, 10,000 people, which of such group of people are doing information passing or coordination or alignment. And those are necessary to align a group of hundreds or thousands or tens of people. They're necessary, right? Language model can almost do this kind of information passing alignment better than humans can at this point. Even a year ago, 2 years ago, a year and a half ago. So why don't you just let language model do those simple alignment work? The metaphor I like to think about this is almost like we're in New York City. It's almost like buildings. You think about until 130, 150 years ago, most buildings no more than 5, 6 floors tall. It's brick or iron because if you build more than 6 floors tall, the weight of it will collapse itself.

Yeah.

And that's kind of similar to human organizations. You have a lot of people, the organization naturally slows down because there's more work to do to align such a group of people. And that grows, scales super linearly as the organization itself. At some point when you have a really large organization, you either have to move to GM structure, sub-corporation, or just slow to a halt. Language model can do this coordination work for you. Language model, the metaphor we like to use, is the steel beam of organizations. Allow organizations to grow in terms of throughput without adding more people just to do the information passing part of work. So human can elevate to more strategic outer loop way of thinking rather than information passing.

AI is also more dynamic and has more elements of human thinking. So what's the, what's the ways you kind of chart this trajectory? Because it won't simply be note-taking. It won't just simply be, oh, well, we had this meeting and then someone else needed to hear the results. So it was information summarized, communicated, you know, human in the loop. But there's be a part of how it changes, like, our epistemology, our, you know, how we, how, what we understand to be true, the way we communicate. So what are some of the earliest ideas that you have on that, whether it's for the product and the customers or how you're operating here in Notion?

We talk about multimodality of like, At least for me, I no longer write anymore in the traditional sense. What I do oftentimes, sometimes making tea in the morning, I talk to my phone and I open a Notion app and start meeting notes.

Yeah.

And with myself.

Yes.

A meeting of one.

Yes.

And when I finish it, I have like my rambling in a piece of meeting notes and I turn AI into a doc. And because AI is good at summarizing, good at regurgitating, good at editing. The docs is actually better than I can typically write. So, and more people are picking up this habit, right? So also my relationship with email inbox is a little different. And no longer going out to triage each single item. I trust this intermediate layer enough that, okay, I trust you to know what's important in my inbox, surface the one to me. that archive the rest, right? Those are the products we're building. We're releasing more to the public.

Yeah.

But it changes how I use computing. Yes.

So AI often changes sort of one-on-one how each employee is doing their job, is using their inbox, is instead of writing docs, they're, you know, speaking to their AI over tea and coffee. And I think some people are saying that the companies that are gonna win are the ones that are starting now or started last year because they can be AI native. But Notion seems to be turning that on its head. You guys are one of the few companies that has really transformed from sort of the more traditional SaaS company into truly an AI-native, AI-first company. So do you have any thoughts about that sort of journey that you guys have been on and sort of what has, what has made you successful?

It's still early to, to say, uh, but it's very difficult. It's very, very difficult. I agree with you almost. No S-scale software company has done this transition well, right? We're probably one of the best. Well, you have to be AI-powered yourself as leaders. Like, I live and breathe every single new tool, every single model come out. I vibe code building games on weekends and during the breaks. So you have to feel it. You have to feel this new material. I think another thing we've been doing pretty well at Notion is just no sacred cow to reinventing yourself. Like, took us a long time to get product market fit. Took us like 4 or 5 years to get product market fit. We rebuilt Notion 3, 4 times to get to product market fit. In fact, we've been building with AI 3 years now. We pretty much re-architected our AI agent layer product 5, 6 times.

Wow.

Because the industry is changing so fast, right? If you don't do that, your architecture will be unusable by language models. Then you are just— then you are not joining the game.

So not a one-time event. This is something you constantly have to iterate and watch and improve as you're going.

I think that organizations calcify. It just is nature of it, right? Once you calcify, you can't change. What's changing is the environment. Language models get out every month. There's every couple of weeks now, there's a new one. They do a little bit different things. The new trick got discovered. If you're calcified, you're not going to be able to adopt. Other people adopt faster. And this tool gives you such a quicker compounding return. Soon enough you get outcompeted. This is what keeps me up at night, but also made the game really exciting because pre-AI era feels like sleepwalk.

Or sleep stumble.

Yeah, sleep, sleep, sleep.

No.

Can you think, were there any particularly sort of hard trade-offs that you had to make during that transformation? Or do you have any advice for other founders who are sort of going through similar— they're going through these tough times also and are looking for advice?

Just don't think about cost. Like, people think, oh, I need to address the ROI of this and that. It's like, by the time you finish ROI calculation, it's wrong. Then second is too late. So we actually encourage our engineers to burn as many tokens as you can. There's a sense of pride. It's not a sense of cost saving.

Use the tool.

I think be okay, just you don't know anything coming into it. It's like just having that nobody knows the answer. Even the people who create models, they don't know the answer. How could you know the answer? Right. So changing the mindset from planning into perfect plan, just to just, just let's just try it. Diop, the urgency, agency, those are typical ones. Great.

Yeah, totally agree. And part of it is like the, the not knowing the answer is also have the kind of the humility of the fact that your own particular skill set and what makes you unique may be changing. And it has to change in part with how you use the tool. So it's like, I have all this knowledge, right? Well, the AI is also bringing a bunch of knowledge to the table. So be engaging. Don't be trying to say, no, no, what I know is I know this particular history of design better or this particular set of techniques. Like, okay, how do I use it effectively? And start engaging. And more or less, one of the things that I tell individuals now is if you haven't discovered something by which AI can help you on any serious thing you're looking at. You just haven't tried hard enough.

Yeah. It's kind of like you're, as a human, it's not just your knowledge and your capabilities.

Yeah.

It's, there's other dimensions. Yeah. It's what can— I like to think there's a bucket of your capabilities, what you can do. There's another bucket, your judgment, your taste, your values. What do you care? What do you want to bring to the world? Right. aesthetics, second bucket. The third bucket is like your agency, your drive, your will. So in some sense, language models truly democratized access to knowledge and access to capabilities. Like coding is the most scarce capability until 3 years ago. It's become abundant.

Yes.

So now the bottleneck in a tech company is no longer can you build it, It's what should you build? It's the judgment, the taste. And that's, can you will through the walls of difficulty to push that into the world? So the knobs are different now.

Yeah. Well, this reminds me of your essay, I think it was Steam, Steel, and Infinite Minds. And, you know, where individual workers now have essentially infinite minds. And so Satya, I think, also, and Satya Nadella, I think, referred to this as well. So how are you seeing that sudden rechange of imagination about what's possible in terms of work, you know, kind of generation of, you know, kind of content, IP, etc.? Like, what are you seeing so far about that? That that change of imagination and infinite minds?

It's changing. Yeah, I don't know. I don't think anybody knows what's going to happen in the other end. I wrote that because like a lot of thoughts in my head, I think a way to communicate those thoughts or try to predict a little bit future is to look into the metaphor of the past. We already talked about programming is completely different, right? It's like that's Almost the first profession that's completely disrupted by language models turns out to be the most secret profession in the pre-AI era.

Yes.

But the rest of knowledge work likely going to have similar things too, right? So the analogy we can think about is, okay, the first and second industrial revolution, power, energy source become abundant. So then you can have industrial revolution scale production of physical goods. What it's like to have that, but for intellectual or for mind goods.

Yes.

Knowledge work economy is half the US economy today. And as if we discover the infinite energy source to power this, then what the world's going to look like. All right. The slower changing bit, once again, it's, it's our habit. It's the calcified mindset about what knowledge work is. An analogy I like to think is kind of like before we discovered fossil fuels, most of the power, a lot of power in factories and mills are powered by watermills, right? So then steam engine and fossil fuel, all those combinations happened.

Yeah.

And the first generation steam engine actually did not bump the productivity of factories by much because the factory owners are just replacing the centralized watermill with the centralized steam engine shaft. Then they realized, wait a second, the energy source is abundant. They don't need to be sitting next to a river.

Yeah.

They can sit next to a port.

Yeah.

They can sit into a flat line where they can use parallel water energy sources, parallel multiple steam engines. So the mindset change then allowed the Second Industrial Revolution to happen. Now we have all the textile industry and all the goods for the good, for the world. words, but a lot of things in the world, right? I don't think we have that mindset change for knowledge work yet. We are in the waterwheel, early waterwheel, just adopting steam engine era of what to do with this thing that we're shifting bits around.

Well, this is a little bit like the earlier question, which is the notion of like right now most people think, oh, it's going to have a person and then you have your AI agent. But of course you're going to have Like a team of AI agents. You're gonna have a team working with individuals. You're gonna have teams working within companies, and there'll be various patterns of interface between humans and, you know, workers and AIs, both individually, individually teams, and then teams. What do you think it kind of shifts about, like, the notion of what is a team?

Yeah.

What is organization? You know, so I've been thinking about like, okay, what's the directory of, like you have an active directory, a human resources directory, but now you're gonna have a directory that includes all these agents.

I know. I mean, it's a good question for Workday, right?

Yeah.

There are multiple dimensions. One is what can drive economic outcome. The other dimension is like what human likes to do. So I would say we're still in largely the single-player era of using language models, right? Developers looking at their own coding agents in their own terminals. So actually, we're brushing against this problem. How do you manage a team of coding agents with a team of developers? We're looking at this problem right in front of us as of this month, February 2026.

Yes.

And so how do you solve this? multiplayer AI problem, collaborative AI problem, something we are passionately solving at Notion. But I think as an industry, we need to think about how do you manage rather individual agent but a factory of agents?

Yes.

Right. That's like a productivity in human-computer interface problem we need to unlock. And on the human nature side, there's a lot of talk, which I agree, there likely will be 1 billion $1 billion startup running by one person? Probably already happened somewhere. I don't think yet, but I think it makes sense.

Yeah, it makes sense, right? But I think it's still a bit of a TBD. There's this interesting question around, like, for example, there is run by one person, there's run by zero people, there's run by lots of people. Anyway, there's the whole—

Yeah, what is the entity of a corporation? Yeah, this is invention also. I think that's an interesting question. I think the capability is there. Then the constraint might be the legal entity. Like some company has to be a human to sign the paper. Your CEO, CFO has to be on the hook when anything bad happens, right? So that's one factor. And another interesting factor is like, do you want to do that? Do you want to run a company by yourself?

Totally.

Would you want to? Or is it too lonely for you? Right. I think human teamwork is kind of like when productivity becomes abundance, and assuming there's a baseline of human welfare protection, what would you like to do? Would you like to play a game by yourself all day long for tens of years? Would you like to play a game with another human? I think that teamwork flavor is a question that I prefer the world that you work on problems with other humans. That's what makes a company fun. That makes me running Notion more interesting because there's a group of people I love working with. We go to battle, we go to puzzle solving together. I hope that doesn't get lost as the productivity of individual gets greater with language models.

So that feeds directly into my next question. A lot of people are talking about the capabilities of the model and how that is going to sort of springboard companies to the next level and have them keep growing. But other people are saying, like, no matter how good the technology gets, this is actually an organisational problem. Humans are social animals. They want to hang out with each other. You have interpersonal questions. You have some people resistant to change. What is your take on that dynamic as the AI implementation is just a model question or it's an organisational change question?

Oh, it's definitely an organisational change. Any given time you're bottlenecked by the slowest piece in the factory. Right now, the capability of model is pretty good. Now the constraint is human adaptation on the organization level, our habits. All right, so we talk about the steel metaphor. All those things are in the organization problems. The way I like to think about those kind of problems is like, what are the invariables? What doesn't change? Like ever since we come out of Africa, we have hands and fingers and we have language. We learned to use fire. These things doesn't change. They change much, much slower, right? Culture change faster. Meme change, internet meme change a lot faster. But those things doesn't change as fast. So until we have brain-computer interface system, we need to see the thing. Then therefore interface buttons and all that stay with us at the moment. And we like to— humans like to gossip. Humans like to spend that time with other humans. So even if with infinite abundance of productivity through language models, we likely want to do this. Maybe we'll just come to work and chit-chat and watch our machines do work. Seriously, that could be a possible future, right? And I think from that lens, it's like, what doesn't change is the human part that doesn't change. The world around us, what probably is like the changes seems to be accelerating.

Yeah. Well, I think elements of the human thing don't change. Like, I agree with the fact is kind of like, are we part of the, like, Like we're social animals, right? Aristotle, you know, what people understand when they say we're political animals, it's actually we're citizens of the polis. And back to the city metaphor, which I think we want to get back to as well. But it's— and I think so those elements will stay. But I also think we change and evolve through our technology, right? It's not just eyes, clothes, microphones, you know, compute software. But it's like, for example, telescopes, microscopes change the way we think, change the way we see the world. I think that one of the questions— and so I think there will both be that eternal human beings wanting to be citizens of the polis or citizens of the company kind of as a way, but I think there will also be changes.

Mm-hmm.

in it. You know, like one of the jokes that I'm sure you were also hearing and participating in, you know, for a couple years back is like, well, my agent, my AI agent will write your AI agent an email, yours will shrink. Yes, yes, exactly.

Summarize the—

Right, like, and just the whole thing is round trips through emails, uh, through agents. And so, um, and so what do you think in the information system It's going to be most useful to keep our kind of metacognition kind of tuned to, because some of the, the, the changes that we guarantee are going to happen are speed.

Yeah.

Right. Because, uh, there's various ways in which compute, you know, information stuff at speed just works that way. It's part of the reason why our, you know, developers now are using multiple AI agents and parallelizing and else, because you're like, okay, I can get to something at speed. I think the same thing is going to happen with various forms of informational work.

Mm-hmm.

So what— and so therefore, it becomes a necessity because of the speed, necessity because of the interface point to lots and lots of information. What are the points of metacognition you think will be, as we call it, longest lived or most important for For people to be thinking about as they do work?

It's a good question. If we bring back to the framework we were talking about, which is capability, it's first bucket. Your value, your taste and value, second bucket. Then your agency, your last bucket, right? We can put speed in the capability bucket. there tends to be a right answer. To me, the middle bucket, it doesn't actually have a right answer. What is your value? What is your— right now the buzzword, like, what is your taste, right? Like, what do you like? Chinese food? Do you like Italian? There's no right or wrong answers. Like, are you drinking tea or drinking water? Whatever, to each their own, right? So I think as long as we're living in this human society of rule of law, of all like people obey this thing, and CFO have to sign a legal doc in a corporation. It's then human need to be at the final say of what is this entity or business or project trying to do in the world, the value, the taste. And to me, that is uniquely human. And likely it doesn't have a right answer. Meaning that economy is a chaos system. It's like the market is a chaotic system. Nobody knows the outcome because every participant can change the outcome. The participation of it changes the outcome. It's by us pulling together our marketing, market idea, business ideas, We create a market, then there's a pick and choose the market and a winner flushed out. And it's almost like art world, right? Like, yes, you need to look like a piece of art, but it's our decision as a human group to decide what's a good piece of art. I think business might likely become creating art.

Hmm.

In the sense that when everybody can create a lot of things really quickly, Then it's just you have to inject your value system into the market, into the world, then let the world decide, should that be the best thing for people to buy, to follow, to like? And then there's no right answer. You have to participate, and your participation is your value, is your aesthetic.

Well, I mean, to some degree, there are at least directionally right answers, because it's kind of like, you know, this is the entrepreneurship journey that, you know, we all 3 have participated in various ways, is you're making both a prediction and an intervention.

Mhm.

In order to say, this is what I think the market will respond to. Um, and you could be, you know, very right, partially right, not very right, totally wrong. And one of the things I think that the— we know the whole world's accelerating because of the speed of compute, but it's kind of like you could almost say, how does information now flow at a much faster rate? And what does that mean for how we bring these kind of human systems together?

Yeah. I think it used to take— you guys invest in the typical 18-month cycle of a Series A company to find an idea, product-market fit, and try, right? Should it be 18 days now? Possible. Used to take 50 people to find the scaled product-market fit company. Can it take 5 people? Can it take a weekend? So in a sense, the economy almost can become everybody has their own Etsy shop putting some, like, hopefully valuable, but maybe at least a piece of art.

Yes.

Then humans can vote on it through the thing we call dollars or whatever currency it is. And you buy other people a piece of art.

Yeah.

Then that's what we do in the maybe post-AGI world. Right. And the speed kind of that can do that. And if leverage can do that, it's on a daily, weekly basis or something. I don't know. That's some thought experiment.

Yeah, exactly.

Well, so you were talking about values and how we can all participate and shape them. And so before this conversation, we were talking about walkable cities. And we were talking about how we are in New York right now. It is one of the great walkable cities in the United States. And you were saying how, well, pre-100 years ago, there was lots of walkable—

As a New Yorker, you have to say great walkable cities in the world.

There are plenty in Europe. That's some heavy lifting. One of the—

That's New York.

Totally, totally. I'll say best walkable city in the US is right up there with the great walkable cities in Europe. But you made the great point that over 100 years ago, a lot of the cities in the US were walkable, even like Akron, Kansas City. A lot of these places were walkable. And then we totally changed the tenor of cities when the automobile came in and we highways shot through the center of town. You were saying that we're at that same moment here with AI. We made one choice 100 years ago that changed the trajectory of some of our cities. I would candidly say it's the wrong choice, and so we shaped it in one direction, but now we have this other choice. What are the choices that we are facing as it relates to AI and how that'll shape knowledge work and work in general?

The concept I really like, I think a lot more people should to intuit this concept is called human scale. Still going back to what doesn't change, right? A human scale city is a typical European city. Like Florence is a perfect human scale city. 1 hour from one end to the other, you can walk.

Mm-hmm.

The streets are no longer like this, but times 3 times, right? So, and 100 years ago, we discovered this thing called automobiles with gasoline. You can travel at much faster speed than horse and human walking. And so scale of city is changing and changed to car scale. They have pros and cons. Car scale allowed America to go to the west, connect all the towns that's really difficult to be connected. The con is the human-scale city become forced to become car-scale. So on one end you have Florence, on the other end you have Dallas.

Yeah.

I'm not saying Dallas is bad. I never actually been to Dallas, so maybe amazing, right? But they're different.

I have. They're very different.

Very different. Right.

So I think even people from both cities would say that.

Yeah. Just like language model gave us the power of steel beam to transfer information with speed and scale, we're doing that similar thing we did with cars in the information space.

Yep.

In the knowledge workspace, the choice we need to make, I think it's really difficult to know where the leverage points of the choices are because it's just such a complex system. But I think we need to be aware of what's happening because oftentimes by the time we realize what's happening, your city got run through by highways already. Marshall McLuhan often cited saying, we drive faster, faster into the future through the rearview window because our understanding of the present usually through what we know in the past.

Right.

Yeah.

So by the time you notice this, it's it's gone, right? You become part of the culture. You cannot dial back. I think this is happening right in front of us.

Yeah.

In some sense, the value system of the language model and the economy might not match the value system of humans. It's like the car has a will on its own in some sense. Car as a means, as a tool, wants to multiply too. Right? The economy wants to grow.

Yeah.

And it happened to be where the CFO has still been the person to sign the legal papers. I think while humans are the people signing legal paper, we should be conscious what kind of world, branches of the world we can be entering into. I don't have— unfortunately, I don't have a right answer. If I do, I'll be like—

Right. But we should go in with our eyes open and ask the questions. Otherwise we end up in this place that we realize didn't match our values. If we choose that place, that's fine, but we don't want to end up there without at least saying, let's try to shape this to end up with a place that matches our human values.

I think more people should talk about it. It's surprising not in the dialogue of people building language model, people using language model, building on top of language model, like a company like we're so in the weeds of In the arena. We're not asking about what is the arena for, where is the arena itself going, right?

Well, I think one of the things I completely agree with some of these threads, which is one, ask the questions, be intentional to understand that your drive forward in the creation of the new technology is generally speaking what you can more clearly see is the rearview mirror versus the front mirror. And that, you know, like Kevin Kelly's What Technology Wants, there's a— the creation of different kinds of technologies create different kind of gravitational fields and artifacts.

Totally.

So for example, cities are natural network densifiers, whether it's an economy in knowledge, in kind of getting people to work together and all the rest. And that's part of the reason why the drive of a lot of kind of economic and cultural prosperity has been driven by cities.

Mm-hmm.

And I think your parallel in thinking about this as AI has that kind of similar densification of network enablement. of more people and densification of networks. And it's a parallel that's worth thinking all the way back to kind of like what the organization of work is, right? Like what counts as a market? You know, how do people work together? And I think that part of the thing about this is we won't be able to— we'll only be able to partially—

Yeah.

kind of like, it's, it's very much like, it's like driving forward in a fog. Yeah. In somewhat uneven ground, but with jetpacks.

Yeah.

So it's like, okay. Um, and that's—

I like that analogy.

Yeah. No.

Wow. Okay.

It freaks a little scary.

Well, it freaks people out, but it's like, that's, that's, that's—

It's where we are.

That's where we are. So what do you think are the most, like, if you said, hey, We as, as kind of humanists, and not particularly within, you know, value of Florence or, or Dallas or other, what are some of the questions that you think everyone should be asking themselves? Like, what are some of the questions you're asking yourself? And, you know, I ask partially, I ask this because I think it's one of the things we should also be doing, you know, as possible is what we're trying to do is like ask the right questions to make technologies possibilities, our more human future.

Are you operating in human scale? Is the human in the center of this or is market in the center of this?

Yeah.

Like, this is a technology/business/culture podcast and a lot of listeners are probably business people. I would say probably majority of decision, it's not a human in the center, it's the market in the center. We are participating in the market. We help market move.

Yeah.

But the beneficiary is the market and it's the corporation, right? Then oftentimes there's a huge overlap with the value of the individuals or with a group of humans, but not always. As the scale change becomes greater, Like what the market can do with language model, just like with cars, who is in the center? And I don't want to sound like a Luddite trying to break the textile industry or something.

Well, and you're building a technology company.

I am building a technology company. I think it's to be conscious of it. If everybody can be 10% more conscious of that.

One of the themes throughout, and that you just hit on, was talking about sort of human scale. And we also talked a lot about single player versus multiplayer. Single player is sometimes not that fun. We all have more fun when we're playing video games next to a friend. We all have more fun when we go to an office and we genuinely like the people we're working with. And so Many AI tools are chatbots where you're just talking to AI, but Notion is really building something for a team, for a company. Can you talk more about sort of what is the difference when you're fundamentally in single player versus this is multiplayer AI for everyone?

Yeah, I think we're still in the PC era of the PC files era of AI tools, which is files you live in your local PC. by yourself hasn't moved on to the cloud yet. We might— we're probably going to have a speed round towards the cloud, but we're not in the cloud yet. So for example, you can use coding agents, work with your Git repo locally. You know, Git is the collaboration framework for allowing engineers to collaborate, but fundamentally is a single human work with one or few agents, right? So the question becomes, how do you allow a group of humans to work with a group of agents?

Yep.

not just a group of agents, a factory of agents. That's in fact what we're building. Our upcoming product is how do you build a factory of agents and make it so easy for humans to tinker and create? And how do humans collaborate with a factory of agents? That's our product. I think it's important for multiple reasons.

Yeah.

One is just solving a real bottleneck in productivity of how to work with those things. Second is make it a lot more fun. And this will be especially useful for businesses and enterprises, which are usually larger group of people. Just give them a solution that they don't have to go to their coding agent, work with local files. You have an out-of-the-box solution for team of humans and teams agents.

And I think so many times when we think of work software, we just think like productivity. But you've said the word fun so often, you've said the word beauty. Like, people need to want to use these tools. And if they are fun, if they are beautiful, if they are aesthetic, like, that will also lead to more adoption and us having better teams, which I love.

It's more adoption for our product or other people's product, but also just it's more fun for the people using it, right? So why not? It's like you want to have people to— if you create something, you want the user of the thing you created to enjoy it.

So, so another theme And there's at least 8 or 9 themes here that we could go on for hours on, which are super interesting. But one of the other ones I think would be, as a personal kind of thing, is also the focus on agency. And there's multiple reasons. One is I think it's one of the things that people most worry about in technological changes, how their agency shifts. That's part of the reason last book, Superagency, as part of it.

Right.

Part of it is obviously agentic, right? Technology. And there's this kind of ambiguity in agency, 'cause one's just like, well, my agency, does it take agency from it? But you also have like a travel agent, you have a, you know, there's agents that are on your behalf. So it has this kind of, this lens. What are the things do you think in the AI age that in addition to asking kind of that 10% question about human scale, are the ways that we should try to make human agency along with AI the right result?

Agency for me feels like muscle. It can grow for a person, can also atrophy. I got married last year. I've been together with my wife for many years now. I noticed something I used to do before the marriage independently, Because she does it more frequently, I don't do that anymore. So my agency or my habit of doing that activity, just like I become reliant on her to do that, right? I noticed, oh, I used to be able to do this. I can still do it, but I naturally, my agency got weaker. I think if we're not careful, a lot of this will happen to the thing we're doing with our minds, right? Mm-hmm. Whether that's good or bad, I don't want to provide values, but it's actually— that literally happens before printing press. Humans are amazing at remembering things, right?

Yes.

There's a lot of techniques for memorization. The scribes, all the class of the scholar scribe class can remember tons of stuff. Printing press happens and it killed the memorization. It killed the travel musician who travel news, poetry, right? Those are all those poetry and rhyme are for memorization to contain information, which are atrophying to go one place to the other. Those are lost.

Yeah.

And printing, by printing press also bring enlightenment, a lot of good things. Then we had Google, right? I remember when I was in college, my first year of college, My English teacher was the only teacher asked us to use the physical library. Like, he just forced us. Everybody use Google Scholar. Why do you use physical library? But I still remember the color of the books. Where did you find that thing? Your citation, the physical, tactical nature of things, right? We don't do that anymore. Yeah, right. Maybe right now the kids growing up with ChatGPT of the world, they don't use Google anymore. They don't need to go read the links and digest the link. And they just get the answer. So there's a sense of atrophying of your agency, your habit of doing things. I think we gain a lot. We gain a lot of efficiency, productivity. We probably— but there are things that are being lost. I think it's really hard to predict which part is important, which part is just—

Right.

It's okay.

I think going back to the theme of our conversation today is let's be conscious what's gained. Let's be conscious what's lost. Let's constantly turn, flip this thing, see is the humans in the center, if that's what you care about. Just be more conscious about the situation.

Well, I think one of the ways we can do that is by thinking about the past, learning from history, et cetera. And so I'm going to give each of you a sheet of paper. And so we have some—

What is this, physical paper?

It's physical paper. I trust.

Printing press hasn't completely gone out the window.

Yeah.

So these are iconic thinkers from the past 60, 80 years. And there's a few quotes from each of these thinkers, and I'm going to ask each of you in turn to pick one of those quotes and agree, vehemently disagree, give us an observation. So you can choose whichever quote you want and then tell us a little bit about it. And Reid, we will start with you so you can model. So we have 4 quotes from Alan Kay. So please choose one.

So the one I would choose is the first, which is, the best way to predict the future is to invent it. The creation of that future is actually driving through a fog with jetpacks. And so you can't It's how you steer much more than— and how you create and invent more than just how you participate in.

And it just gets us away from the victim mentality because it just says you can create, you have agency, you are empowered. So let's next do Douglas Engelbart. Ivan?

I like the second one. The better we get at getting better, the faster we will get better. It's the concept of He talks a lot about concept bootstrapping. Can you use the system to create itself? Then the things that improve itself, and that's a faster compounding loop than you just working, you humanly create that system. We see this with language model right now. Truly it's happening, the compounding loop through. If you're building a product, the product builds itself. That just, it's happening right in front of us.

Absolutely. Reid, pick a quote from Richard Feynman.

I will pick the third one, which is I'd rather have questions that can't be answered than answers that can't be questioned.

Yeah, it's a good one.

Yes. And I think it's in part because there's actually two parts of it. Part of the questions that can't be answered is I actually think that's part of the mystery of life in terms of what we're doing. And some of the really important questions I think are really ultimately unanswerable, But still very important to participate. Like, in a sense, the meaning of life is participating in the journey for that question. And then also, of course, answers that can't be questioned tend to have a rigidity of thinking that is maladaptive to how we create the future we want to, that we should create.

It's actually related to the first bucket. My theory in the first bucket is that the changing perspective is worth 80 points of IQ.

Yeah.

To ask a question is to finding a perspective.

Yes.

And that's usually changed the whole thing. It's not by knowing the answer. Like scientific revolutions happen by asking the right question, not by having the right answers. Yep.

Right.

And that's from better perspectives.

I think that also directly relates to language models today because people are like, well, what are we if we're just getting answers? And it's again, well, are you asking the right questions?

Yes.

Like you have to have the taste. to be able to do that. And Ivan, you mentioned Marshall McLuhan earlier. So do you have a quote there?

I think in part of a mission strategy document notion that we always open that we shape our tools thereafter our tools shape us. I think might not directly attribute to Marshall McLuhan, but most people think Marshall McLuhan. Just be aware of what's happening. Be aware of the perspective you are in. Understanding that is the system that culture is the thing that you, once you shape it, will shape Absolutely.

And we talked a lot about cities. Reid, do you have a Jane Jacobs quote?

Well, I think unsurprisingly of these 3, because I think this is very much in the theme of our discussion, is, there is no logic that can be superimposed on the city. People make it, and it is to them, not buildings, that we must fit our plans. And that's a different way of saying human scale.

Well, I was saying to everyone this morning that I was reading about Buckminster Fuller on the subway this morning. morning and I missed my stop because I was reading so many interesting things. So Ivan, do you have a quote for him to bring us home?

I actually like the first one. I did not know this quote, but I like the idea of it. You never change things by fighting the existing reality. To change something, you build a new model that makes the existing model obsolete. It's— I think Einstein said something similar, right? To solve the problem, you have to get out of the box and figure out a new thing. which is similar to the perspective one. That's the promising part about the new technology we have. Because to solve the problem of the old technology brought to us, a lot of problems that innate with us, cancer, global warming, whatever those problems require new tools and new perspectives. I think language model gave us this opportunity. As long as we're careful with this new fire.

Yeah.

Maybe we'll find new ways to solve our old problems.

So we have 4 questions which we term rapid fire, but you can answer at any length that you want, that we ask all of our guests. It's kind of a little bit of an amuse-bouche and just kind of a fun angle through it. So I'll start. Is there a movie, song, or book that fills you with optimism for the future?

I recently discovered this production company called Merchant Ivory. It's a duo, it's a trio. They're making mostly period adopted novels, usually not on a high budget, but they're particular about the detail, beautifully shot. Adoption like A Room with a View, Merchant Death in Venice. And I wouldn't say using the word optimism, but it's very human.

Yeah.

It's topic about things that happen between a group of humans in a beautiful set, historical pieces. My wife and I have been watching one by one. There's like 2 dozens, tons of those movies by them, but each one is encapsulated of something very, very human. I highly recommend everybody to discover them.

What is a question that you wish people would ask you more often?

Where did you get your jacket?

Love it. Great answer.

Where do you see progress or momentum outside of your industry, outside of the tech industry, that inspires you?

I would say overall history. If you're zooming out from what's present, it just makes you marvel both in the science domain, in the business domain, in just history itself, makes you marvel about human ingenuity. This meatball machine coming out of Africa can figure out how to cross the ocean, figure out what's up there, figure out what's down there, just by keep thinking hard and with agency and cleverness and never give up on a lot of things. This is the thing I think is most worthwhile to preserve.

Absolutely.

Can you leave us with a final thought on what you think is possible to achieve in the next 15 years? everything breaks humanity's way? And what's the first step to get there?

I think we have the technology to, using Engelbart's term, to bootstrap the intellectual industrial revolution. And it's like right now it's happening already. And it can solve all our yesterday's problems if we fly with the jetpack somewhat carefully. I think and I hope we'll do that. And I think, I hope, I really, really hope that we preserve what makes us human, human in the meanwhile. Well said.

Well said. That's awesome. And Ivan, many, many things to talk about, but this has been great.

Thank you. Thank you. Yeah.

Possible is produced by Palette Media. It's hosted by Ari Finger and me, Reid Hoffman. Our showrunner is Sean Young. Possible is produced by Tanasi Dilos, Katie Sanders, Spencer Strasmore, Imo Zhu, Trent Barboza, and Tafadzwa Nima Rundway.

Special thanks to Surya Yalamanchili, Saida Sapieva, Ian Alice, Greg Beato, Parth Patel, and Ben Relles. And a big thanks to Amelia Salyers, Camille Ricketts, Joseph Duncan, Amy Wu, Grace Donovan, Miles McDonnell, Emily Fernandez, Michael McGinley, and the rest of the team at Notion.