Ivan Zhao, CEO of Notion, discusses how his company transformed from a traditional SaaS platform into an AI-native organization with over 700 AI agents working alongside 1,100 employees. The conversation explores the philosophical implications of AI as a new material for building software, drawing parallels between the current AI revolution and historical technological shifts like the industrial revolution and the rise of automobiles.
- AI transformation requires constant re-architecting - Notion rebuilt their AI agent layer 5-6 times in three years as the technology evolved rapidly
- The bottleneck in tech companies has shifted from 'can you build it' to 'what should you build' - judgment and taste become more valuable than coding ability
- Organizations need to consciously choose between human-scale and machine-scale systems, similar to how cities chose between walkable designs and car-centric infrastructure
- Language models can serve as the 'steel beam of organizations' - enabling coordination and information passing at scale without adding human overhead
- The transition from single-player AI tools to multiplayer AI systems represents the next major evolution in workplace productivity
"Language model can almost do this kind of information passing alignment better than human can at this point"
"The bottleneck in a tech company is no longer can you build it, it's what should you build"
"We are in New York City. It's almost like buildings you think about until 150 years ago, most buildings no more than five, six floor tall. It's brick or iron. Because if you build more than six floor tall, the weight of it will collapse itself"
"Just don't think about cost. Like people think, oh, I need to adjust the ROI of this and that. Like by the time you finish ROI calculation it's wrong then seconds too late"
"You never change things by fighting the existing reality. To change something, you build a new model that makes the existing model obsolete"
In 2015, Ivan Zhao and his co founder moved from San Francisco to Kyoto, Japan, where they spent 18 hours a day in a two story house so small that only a traditional shoji screen separated their bedrooms. Rebuilding Notion from scratch. After scrapping three 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.
0:00
Today, Ivan runs a company with over 700 AI agents, working alongside roughly 1100 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.
0:27
Today, we're, 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?
0:50
Artistic sensibility. That's it. It's very serious.
1:43
Yes. We can use less. Less serious work.
1:45
Oh, no, no. This is my artists. So it's kind of intuitive, I would say. Just like since Notion, we're small, we all 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, so why not surround yourself with timeless office tools so we can design timeless software?
1:48
Yeah, no, makes sense. So like, one of the things I heard about is apparently there was a rug from your childhood. Yes, that was in Notion's first office.
2:23
So it was until like our last office. Until maybe six months ago.
2:32
Okay.
2:36
I grew up in Iraq from My hometown, because my hometown was in this Muslim part of China. So we use a lot of rugs. Just there is this red and blue rug we've been using ever since the first Notion office and has been travel with us to all the offices. And six months ago we moved to a new office in sf. So I finally took the rug home. Yeah, it's still clean.
2:38
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?
3:03
A lot of people think Notion is the productivity tools, AI power 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. The hippie generation took asset thought about okay on the west coast, largely, 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 do so. I've been working on Notion as idea more than a decade. Started as this kind of you can do everything product. In the last three years, AI is this kind, you can do everything new technology. So it's become a new piece, a new Lego block in the two set for us.
3:14
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. 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?
4:30
At least the first model that clicked with us, with me, was the GPT4 class. GPT 3.53 class is like useful, but if you see GPT4 class model, it's like it's not just a text regurgitator.
4:52
Yes.
5:07
It's a piece 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 GPT4 because our friends were working on 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. The story was my co founder Sam and I got access during a company retreat in Mexico, Cancun, and we just lock ourselves in the hotel room. The entire company retreat except the Kinoa happy due. Then we're just building the first prototype. And if you build tools with this, you know, it's a completely different material.
5:07
Yeah, no, I remember the GPT4 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 book before last one impromptu to try to get. To show AI as a collaborator in writing a book, that piece of the human mind as ways of doing it. Because as you probably remember, the GPT4 launch date was kind of moving. So we ultimately ended up publishing it through Amazon directly to be able 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 GPT4. Now when you think about Notion has this kind of textual workspace, but part of what, you know, brings up in 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 that notion's becoming and adding that in?
6:09
Yeah. Rather than consider as a fabric, we like to think in terms of building blocks. And to me it's very difficult for one company to change the building block or language of anything that humans use. Software UI is kind of just like a language. It's like we all grew up using graphic user based interface. We understand there's a box, you're supposed to click on it, double click mean a different thing. It's almost like speaking French or English. It's a language you learn and there's all the building block within a language in some sense. The premise of the first version of Notion was okay, what are the core language of computing historically always been in the hand of programmers. Can we open up to allow non programmer to stitch it together those building blocks? The building block being text you mentioned in one text editing relational database. It's 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 open 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 we sort of can see work. 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 Chatbot? Google. It's a chatbot. It's a text input.
7:33
Yes.
9:40
That's how we understand this new tech. The most powerful new technology we have language model. We mirror looking the image of Google. Right. And if you think about in the past couple years, year or so coding agent become really popular. Id become popular. So everybody was running coding agents. But in the past six, nine months people realize you have a limited bandwidth managing one coding agent. When you're managing a dozen coding agents, what do you do? It's a freaking Kanban board. We're going back to a project management software. So 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.
9:40
How is the kind of the 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?
10:25
Yeah, I don't think anybody knows. I think that's, that's the honest answer. Yeah, Nobody know the right answer. Everybody share practice on Twitter, on blog posts. So figuring out this together, I would say the overall trend is if you think you have a company of ten people or a hundred people, or a thousand people, ten thousand people. Which of such group of people are doing information passing or coordination or alignment? And those are necessary to align a group of hundred or thousand or tens of people. They're necessary. Language model can almost do this kind of information passing alignment better than human can at this point, even a year ago, two years, 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 150 years ago, most buildings no more than five, six floor tall. It's brick or iron. Because if you build more than six floor 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 slow down because there's more work need to do to align to such a group of people. And that growth 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.
10:39
AI is also more dynamic and has more elements of human thinking. So 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. Those information summarized communicated human in the loop. But there's be part of how it changes our epistemology, 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?
12:37
In notion, we talk about multimodality, 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 with myself. A meeting of one. When I finish it, I have my rambling in a piece of meeting notes and I turn AI into a doc. Because AI is good at summarizing, good at regurgitating, good at editing the doc is actually better than I can typically write. So. And more people are picking up this habit. Right. So also my relationship with email inboxes was 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 but it changes how I use computing.
13:22
Yes.
14:29
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 speaking to their AI over, over tea and coffee. And I think some people are saying that the companies that are going to 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?
14:30
It's still early to say 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 pelt yourself as leaders. Like I live and breathe every single new tools, every single model come out. I vibe, code building games or on weekends on during the break. So you have to feel, you have to feel this new material. I think another thing we've been doing pretty well at Notion is just no secret cow to reinventing yourself. Like took us a long time to get product market fit. Took us like four or five years to get product market fit. We rebuilt Notion three, four times to get to product market fit. In fact we've been building with AI three years now. We pretty much re architect our core AI agent layer product five, six times. Wow. Because the industry changes so fast, right. If you don't do that your architecture will be usable by language models, then you are not joining the game.
15:08
Not a one time event. This is something you constantly have to iterate and watch and improve as you're going.
16:24
I think that organizations calcify the nature of it, right. Once you calcify you can change. What's changing is the environment, language, model. Get out every month, every couple of weeks now there's a new one. They do a little Bit different things. The new trick got discovered. If you are calcified, you're not going to be able to adopt other people adopt faster and this will give you such a quicker compounding return. Soon enough you get out competed. This is what keeping me up night but also made the game really exciting because pre AR era feels like sleepwalk or sleep stumble.
16:29
Yeah.
17:09
Sleep, sleep. No.
17:10
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.
17:11
Just don't think about cost. Like people think, oh, I need to adjust the ROI of this and that. Like by the time you finish ROI calculation it's wrong then seconds too late. So we actually encourage our engineer to burn as many tokens as we can. Yeah. This is a sense of pride, it's not a sense of cost saving.
17:24
Yeah.
17:44
Use the tool. I think the. Okay, just you don't know anything coming into it. It's like just having that nobody know the answer. Even the people who create model, 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. Die up the urgency agency. Those are typical ones. Great.
17:45
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 table. So be engaging. Don't be trying to say no. No. What I know is I know this particular, you know, 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.
18:08
Yeah, it's kind of like as a human, it's not just your knowledge and your capabilities.
19:00
Yeah.
19:06
It's. There's other dimensions. Yeah. It's what can. But I like to think there's a bucket of your capabilities, what you can do, there's another bucket. Your, your judgment, your taste, your values, what do you care? What do you want to bring to the World. Right. Your aesthetics second bucket, the third bucket is like your. Your agency, you drive, your will. So in some sense language model truly democratized access to knowledge and access to capabilities like coding. It's the most scared capability. Until three years ago it become abundant.
19:06
Yes.
19:43
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 can you will through the walls of difficulty to push that into the wall, into the world. So. So the knobs are different now.
19:44
Yeah.
20:00
Right.
20:01
Well, this reminds me of your essay I think was steam steel and infinite M 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 re. Change of imagination about what's possible in terms of work, you know, kind of generation of you know, kind of content, ip, et cetera. Like what do you. What are you seeing so far about that? That. That change of imagination and infinite minds
20:01
as changing? Yeah, I don't know. I don't think anybody know what's going to happen in the other end. I wrote that because 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. I already talk about programming is completely different almost the first profession that's completely disrupted by language model turns out to be the most secret profession in the free era.
20:45
Yes.
21:18
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 is like to have that before intellectual or for mine goods? Yes, Knowledge war economy is half the U.S. economy today. And as if we discover the infinite energy source to power this, then what the world's going to look like the slower changing bit. Once again it's our habit. It's the calcified mindset about what knowledge world is. An analogy I like to think is kind of like before discover fossil fuels. A lot of power in factory and mills are powered by water mills. So then steam engine and fossil fuel, all those combinations happen. And the first generous steam engine actually did not bump the productivity of factory by much because the factory owners are just replacing the centralized water mill with a centralized steam engine shaft. Then they realize wait a second, the energy source is abundant. They don't need to be sitting next To a river, they can sit next to a port, 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 happened. Now we have all the textile industry and all the goods for the good, for the worst. But a lot of things in the world. Right. I don't think we have that mindset change for knowledge for yet. We are in the water wheel, early water wheel, just adopting steam engine era of what to do with this thing that we're shifting bits around.
21:19
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 going to 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 individual teams and then teams. What do you think it kind of shifts about like the notion of what is a team.
23:15
Yeah.
23:49
What is organization? You know. You know, so I've been thinking about like, okay, what's the directory of? Like you have a, an active directory, human resources directory. But now you're gonna have a directory that includes all these agents.
23:50
I know, I mean it's a good question for workday, right?
24:04
Yes.
24:07
There are multiple dimensions. One is what can drive economic outcome. The other dimension is like what it's what human likes to do. So I would say we're still in the largely the single player era of using language model. Right. Developer looking their own coding agents in their own terminals. So actually we're brushing against this problem. How do you manage a team of coding agent with a team of developers? We're looking at this problem right in front of as of this month, February 2026. And so how do you solve this multiplayer AI problem? Collaborate AI problem, something we are passionately we're solving at notion. But I think as an industry we need to think about how do you manage in rather individual agent but a factor of agent.
24:08
Yes.
24:54
Right. That's like a productivity inhuman computer interface problem to unlock. And on the human nature side, there's a lot of talk which I agree there likely will be 1 billion people, $1 billion startup running by one person. Probably already happened somewhere, I don't think yet. But I think it's makes sense.
24:54
Yeah, it makes sense.
25:17
Right.
25:18
But I think it's still a bit of A tv. 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, this is the whole.
25:18
Yeah. What is the entity of a corporation? It's an invention also. I think that's the 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? Is it too lonely for you? Right. I think human teamwork is kind of like when productivity become 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 flavors is a question that I prefer the world that you work on problem with other humans that 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.
25:31
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 no matter how good the technology gets, this is actually an organizational problem. Humans are social animals. They want to hang out with each other. You have have interpersonal questions. You have some people resistant to change. So like where do you what is your take on that dynamic as sort of the AI implementation is sort of just a model question or it's an organizational change question.
26:42
Oh, it's definitely organizational change. I think any given time you're bottomed at 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. 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 learn to use fire. These things doesn't change. They change much, much slower. Culture change faster. 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 human like to gossip human like to spend that time with other humans. So even if we have infinite abundance of productivity through language models, we likely want to do this. Maybe we just come to work and chit chat and watch our machine. Seriously, that could be a possible future. Right? And I think from that lens it's like what doesn't change? It's the human part that doesn't change. The world around us were probably. It's the changes seems to be accelerating.
27:17
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 they're back to the city metaphor 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 closed microphones, you know, compute software, but it's like, you know, 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 they will both be that eternal like human beings wanting to be, you know, citizens of the police or citizens of the company kind of as a way. But I think there will also be changes 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 and email. Yours will shrink. Yes, yes, exactly. Like in just the whole thing is round trips through emails, through agents and so. 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 changes that we guarantee are going to happen are speed.
28:51
Yeah, right.
30:35
Because 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 also 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, 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 I call it, longest lived or most important for. For people to be thinking about.
30:36
Yeah.
31:18
As they do work?
31:18
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. Yeah. There tend 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 ran out the buzzword. Like what is your taste?
31:19
Right.
31:54
Like what do you like Chinese food? Do you like Italian? There's no right, there's no right or wrong answers. Like are you drinking tea or drinking water, whatever. To each of 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 to me that is uniquely human and likely doesn't have a right answer. Meaning that economy. It's a chaos system. It's like the market is a chaos chaotic system. It's nobody know the outcome because every participant can change the outcome. The participation of it change 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 when they're flushed out and it's. It's almost like, 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 decide what's a good piece of art. I think business might likely become creating art 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, it's your aesthetic.
31:55
Well, I mean to some degree there are at least directionally right answers because it's kind of like this is the entrepreneurship journey that we all three have participated in various ways. Is you're making both a prediction and intervention in order to say this is what I think the market will respond to. And you could be you know, very right, partially right, not very right, totally wrong. And one of the things I think that the like, you know, we know the whole world's accelerating because of the speed of compute but it's kind of like you could almost say like, you know, how does information now flow at a much faster rate and what does that mean for you know, kind of how we bring these kind of human systems together.
33:50
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 company. Can it take for 5 people? Can it take a weekend? So in a sense the economy almost can becomes everybody has their own Etsy shop putting some hopefully valuable but maybe at least a piece of art. Then humans can vote on it through through the thing we call dollars or whatever currency it is and you buy other people a piece of art, then that's what we do in the maybe post AGI world. Right. And the speed can do that and if leverage can do that is on a daily weekly basis or something. I don't know. That's some style experiments.
34:45
Yeah, exactly.
35:41
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
35:42
lots of walkable as the New Yorker you have to say great walkable cities
36:00
in the world in that one, plenty. In the Europe I have some heavy
36:03
lifting one of the New York. Totally, totally. I'll say best walkable city in the US Right up there 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 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 highways shot through the center of town. And you were saying that we're sort of at that same moment here with AI we sort of made one choice 100 years ago that sort of change 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. And so what are the choices that we are facing as it relates to AI and how that'll shape knowledge work and work in general.
36:09
The concept I really, really like. I think a lot more people should intuitive. This concept is called human scale. Still going back to what doesn't change. Right. A human scale city is a typical European like Florence is a perfect human scale city. One hour from one end to another you can walk the street are no longer like this but times three times. Right. Hundred years ago, we discovered this thing called automobiles. With gasoline you can travel at much faster speed than horse and human walking. So scale of cities changing and change to car scale. They have pro and cons car scale allow 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. I'm not saying Dallas bad. I've never actually been to Dallas, so maybe amazing. But they're different. I have.
36:58
They're very different.
38:12
Very different.
38:13
I think even people from both cities would say that.
38:14
Yeah. I think just like language model give us the power of steel beam to transfer information and and speed and scale. We're doing that similar thing we did with cars in the information space, in the knowledge work space. 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. Martin McCun often cited saying we drive faster, faster into the future through the rare view window because our understanding of the present usually through what we know in the past. So by the time you notice this, it's gone. You become part of the culture. You cannot dial back. I think this is happening right in front of us.
38:16
Yeah.
39:19
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 meme, as a tool wants to multiply too. Right. The economy wants to grow. And it happened to be where the CFO has still be the person to say sign the legal papers. I think while human are the people signing legal paper. We should be conscious what kind of world branches of world we can be entered into. I don't have. Unfortunately, I don't have a right answer. If I do, I'll be like, right.
39:21
But, but we should, 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.
40:02
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 company. We're so in the weeds of, in the arena. We're not asking about what is the arena for. Where does the arena itself going.
40:18
Well, I think one of the things, I completely agree with some of these threads, which is one, ask the questions, be intentional. Two, understand that your, your drive forward in the creation of the new technology is generally speaking, what you can more clearly see is the rear view mirror versus the front mirror. And that, you know, like Kevin Kelly, is what technology wants. There's a, the creation of different kinds of technologies create different kind of gravitational fields and artifacts.
40:42
Totally.
41:15
So, for example, cities are natural network densifiers, whether it's an economy in knowledge, in, you know, 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. 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, like what counts as a market. You know, what, you know, how do people work together? And I think that part of the, the, the thing about this is we won't be able to. We'll only be able to partially like, kind of like it's, it's very much like. It's like driving forward in a fog in somewhat uneven ground, but with jet packs. Yeah. So it's like. Okay. And that.
41:16
I like that analogy.
42:29
Yeah.
42:30
Wow. Okay.
42:31
It freaks.
42:32
A little scary.
42:32
Well, it freaks people out. But it's like that, that's, that's, that'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 asked partially 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 are more human future.
42:33
Are you operating in human scale? Is the human in the center of this or it's market in the center of this? Like this is a technology business slash culture podcast and a lot of listener probably are 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, we're participating in the market. We help market move. But the beneficiary is the market. It's the corporation. 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 become greater, the market can do with language model just like with cars. Who is in the center? I don't want to sound like trying to break the textile industry or you're
43:12
building a technology company.
44:25
I am building a technology company. I think it's to be conscious of it if Everybody can be 10% more conscious of it.
44:26
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.
44:38
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're probably going to have a speed run 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. Git is the collaboration framework for allow engineer to collaborate. But fundamentally is a single human work with one or few agents. So the question become how do you allow a group of human to work with a group of agents, not just a group of agents or factors of agents. That's in fact what we're building. Our upcoming product is how do you build a factory of agent and make it so easy for human to tinker and create and how do human collaborate with a factory agent? That's our product. I think it's important for multiple reasons. One is just solve 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 business enterprises which are usually larger. 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 out of the box solution for team of human and teams agents.
45:13
And I think so many times when you think of work software we just think like productivity. But you've said the word fun so often. You've said the word beauty. 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.
46:37
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 help people to. If you create something, you want the user of thing you created to enjoy it.
46:55
So another theme, and there's at least eight or nine themes here that we could go on for hours on, which is super interesting. But one of the other ones I think would be like as a personal kind of thing is 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, Super Agency is as part of it. Part of it is obviously agentic, right? Technology. And there's this kind of ambiguity in agency because one, just like 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, 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?
47:09
Agency for me feels like muscle. It can grow for a person, can auto 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 in 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 I could become rely 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 mind. Right. Whether that's a good or bad. I don't want to provide values. But actually that literally happens before printing press. Humans are amazing and remember things. Right. There's a lot of techniques for memorizations. The scribes. All the class of the scholar scribe class can remember tons of stuff. Printing press happens. It killed the memorization. It killed the travel musician who travel news poetry. All those poetry and rhyme are for memorization. Contain information which are atrophying to go one place to the other. Those are lost by printing press. Also bring enlightenment. A lot of good things. Then we had Google. 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. 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 those things? Citation. The physical tactical nature of things. We don't do that anymore. 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. Yeah. And they just get the answer. So there's a sense of atrophi of your agency or 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 in gain. Let's be conscious what's lost. Let's constantly turn flip this thing. See the humans in the center. If that's what you care about. Just be more conscious about the situation.
48:14
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.
51:01
What is this? Physical paper.
51:12
Physical paper. I trust
51:13
printing press hasn't completely gone out the window.
51:17
Yeah.
51:20
So These are iconic thinkers, you know, from the past 60, 80 years. And I'm gonna. There's a few quotes from each of these thinkers, and I'm gonna ask each of you in turn to pick one of those quotes and agree.
51:21
Vehemently.
51:36
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 four quotes from Alan Kay, so please choose one.
51:36
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 do you participate in.
51:49
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.
52:09
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 improving 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. It's happening right in front of us.
52:20
Absolutely. Reid, pick a quote from Richard Feynman.
52:49
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.
52:55
Yeah, it's a good one.
53:02
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. In a sense, the meaning of life is participating in the journey. For that question, 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 that we should create.
53:03
It's actually related to the first bucket. My theory in the first bucket is the changing perspective is worth 80 points of IQ. Yeah. To ask a question is to finding a perspective.
53:40
Yes.
53:52
And that's usually changed the whole thing, it's not by knowing the answer, like scientific revolution happened by asking the right question, not by having the right answers. Right. And that's from better perspectives.
53:52
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? Like, you have the. You have to have the tape to be able to do that. And Ivan, you mentioned Marshall McLuhan earlier. So do you have a quote there?
54:03
I think in part of a mission strategy doc in notion, 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 it's the system, that culture is the thing that once you shape, it will shape back to. Absolutely.
54:18
And we talked a lot about cities. Reid, do you have a Jane Jacobs quote?
54:44
Well, I think, unsurprisingly of these three, because I think this is very much in the theme of our discussion is there is no logic that can be superimposed in the city. People make it, and it is to them not buildings. We must fit our plans. And that's a different way of saying human scale.
54:48
Well, I was saying to everyone this morning that I was reading about Bunkminster Buckminster Fuller on the subway this 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?
55:07
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. 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, we brought to us a lot of problem 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 science were careful with this new fire. Yeah, maybe we'll find new way to solve our old problems.
55:19
So we have four questions which we term rapid fire, but you can answer at any length you want that we ask all 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?
56:07
I recently discovered this production company called Merchant Ivory. It's a dual. It's a trio. They're making mostly period adoptive 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 and I wouldn't say using the word optimism but it's very human. It's topic about things that happen between a group of human in a beautiful set. Historical pieces my wife and I have been watching one by one. There's like two dozens, tons of those movies by them but each form is encapsulated of something very, very human. I highly recommend everybody to discover what
56:27
is a question that you wish people would ask you you more often.
57:19
Where did you get your jacket? Love it.
57:22
Great answer.
57:26
Where do you see progress or momentum outside of your industry side of the tech industry that inspires you?
57:27
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 make you marvel about human ingenuity like 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.
57:33
Can you leave us with a final thought on what you think is possible to achieve in the next 15 years, if any. Everything breaks humanity's way. And what's the first step to get there?
58:10
I think we have the technology to using Engelbart's term to bootstrap the intellectual industrial revolution. And like right now, it's had happening already and it can solve all our yesterday's problem 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.
58:18
Well said, well said. That's awesome. And I have many, many things to talk about but this has been great.
58:47
Thank you. Thank you. Yeah.
58:52
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 Tenasi Delos, Katie Sanders, Spencer Strasmore, Emo Zhu, Trent Barboza and Tafadzwa Nima Rundway.
58:54
Special thanks to Surya Yalamanchili Sayda Sepieva, Ian Alice, Greg Beato, Parth Patil and Ben Rellis. 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.
59:09