In Conversation with Ivan Zhao

On New Year's Day 2021, Notion crashed because you were running out of database storage, and you ended up pausing new features for six months to rebuild the infrastructure. What were you saying to yourself in those first ten minutes when you realized this had happened?

We knew it was coming, first of all. There's a doomsday clock. You can see, okay: how long does it take to vacuum your Postgres database? And it becomes longer and longer and more sporadic. We were running out of the largest Postgres instance you can run on AWS. So there's no other way out besides sharding the database.

I think our engineering philosophy has always been more product-focused and pragmatic, so we didn't optimize for it prematurely. But we went to the other extreme and did it way too late. At the time, it was all hands on deck. Our engineering lead at that moment did a good job with a couple of key team members. He was like, "Let's stop everything. If this goes down, that means all Notion users can only read their Notion content. They cannot write." That would basically end our business.

So we just did nothing for a while to get this back on track. It's very critical. We have a doomsday clock every day.

And you still have the clock today?

Right now, we've repurposed the clock to show how many days we've been up without severe server issues.

Okay. So it's a more positive spin on the same clock.

Used to be counting down, now it's counting up.

From the outside, it looks like there have been a couple of reset moments in the company's history—one in 2015 and perhaps another in 2021. Would you call those resets?

I wouldn't call the 2021 one a reset. COVID happened; nobody knew how to react. We did our best. It wasn't a sudden change in strategy or worldview.

The Japan one in 2015 was because we were literally going to run out of money if we didn't do something hardcore. We were at this place where we were evaluating: what do we care about? We looked up to companies like IKEA, like Uniqlo — companies that bring quality products to a lot of people. We wanted to do the same with software.

For the longest time, software has been the scarcest product resource in the knowledge-work economy. So we spent many years trying to figure out: how do we democratize software tool-making? But that's a very open-ended problem. We spent two or three years trying to find a product for this. We tried to build two, three, four different things. No customers. Nobody cared.

Eventually, we stumbled upon a form factor kind of like today's Notion. But we were running out of money building it with a team of five-ish people, and we already had a lot of tech debt. We needed to rebuild from scratch. We cut back to two people. Morale was low, so we decided to go somewhere neither of us had been, just to change our mood a little bit. We went to Japan.

How confident were you at the time? How did you distinguish between “we're early, we just need more time” and “this simply isn't working”?

We were quite confident. We knew that the thing we could see—the new version of Notion back then—could be built, and our capability kept getting better.

We just needed to try again and build a version that was production-ready. We knew we could do it, and we knew once it was done, it would be better than all the products on the market. So we actually weren't too panicked about this.

Eventually, we did run out of money. My mom had to lend me money.

And that bridged you until the next funding round?

Yeah. Pretty much.

Were you already charging customers at that point, or was the usage enough to give you confidence that the product itself was working?

We had an older version of the product that was similar to the new version we were building. But it was technically unreliable. Even though people said good things about the user experience and we could see product-market fit, the reliability wasn't there.

So the question wasn't really whether people wanted it anymore. It was whether you could make the technology live up to what you already knew the product could be.

I think Simon and I are both pretty even. Our highs are never that high, our lows are never that low. There's a Chinese saying: "Newborn calves are not afraid of tigers." They don't know what tigers are.

I wonder how much of this predates Notion. You moved around a lot growing up, then moved to Vancouver when you were 16 or 17. What was that move about?

My mom cared about education. She wanted to move to a Western country.

Were you against it at the time—new country, new language, starting over again?

No. I grew up in resets all the time. I just followed my mom around China. I went to probably four different elementary schools, so I was pretty used to that.

You seem to have always had these two sides: art and programming. I understand you won an art contest as a young kid and later became an international programming Olympiad champion. Where do you think those two interests came from?

I’m not sure. My parents tried to support my interests, but not force anything. In China, you usually have to pick an instrument—piano or violin, typically. I never picked an instrument. I think my interests pushed me toward watercolor painting or Chinese calligraphy. I was drawn to that.

And programming?

Programming is just so close to playing video games. I’m good at math. I’m good with programming. You need to win some competitions to get into a good school. So of course I picked the one that allowed me to play with my computer.

You said you were a college senior when you became interested in the history of computing. I don’t think it’s that common as a college student to be inspired by the technology and ideas of the previous 60 years. What do you think drew you to that history?

I don’t know, but it’s true. I studied cognitive science, which is adjacent to computer science. Most computer scientists don’t know who the pioneers are. It’s as if you’re studying physics and you don’t know who Isaac Newton is. You should get fired.

What were you trying to bring back from that earlier era of computing?

The essence is that computing should be a medium that’s rather malleable. A non-programmer can modify it. There’s no real boundary between the software user and software maker. Not everybody needs to make or tinker with their own tools, but if you want to, there’s nothing stopping you.

Early computing systems had it. Smalltalk had it. HyperCard sort of had it. The nineties generation lost it. So we’re really trying to bring this back.

If we had that, SaaS wouldn’t have happened. SaaS happened because spreadsheets aren’t powerful enough to support more complex workflows, and software is too complex to build. The gap between spreadsheets and enterprise software is SaaS, right? So we’re trying to fill that gap—consolidate all the SaaS tools into a flexible, Smalltalk-like system.

Before you arrived at something that looked like today’s Notion, you said there were three or four different versions of the idea. What did those look like?

The first version was kind of like Figma meets Webflow. You could build apps; you could build websites. But people didn’t know: should I build websites with it? Should I build apps with it?

Then there was the developer tool. It was kind of like a more powerful version of Firebase. Those were the two big directions we went really deep on.

Eventually, we ended up with productivity software on the outside, but a visual programming environment on the inside.

Is it fair to say that Notion started to work with customers as you abstracted away more of the visual programming environment?

I think the thing we're doing well is that we're both computing nerds but also good with design. Smalltalk sort of failed because the design is revolutionary, but it's not easy to pick up. So how do you build the bridge? We're good at building bridges.

Some of this comes back to what Xerox PARC was. It was a place that could have no commercial intent, and that cut both ways. Other companies commercialized its inventions, but you also wonder whether the lack of commercial intent gave them the space to invent so much.

Today, there’s a return of the research lab. Do you think that middle period—the SaaS era—was shaped by the overly commercialized nature of computing?

I think that’s probably true. The history of Xerox PARC is, the DARPA people—a bunch of them went to Stanford and worked with Engelbart, who did the “Mother of All Demos,” and somehow ended up at Xerox PARC. It was the same group of people.

You can argue the AI era started with the research community. Like DeepMind’s Demis Hassabis—he preferred to cure cancer and discover the science before any commercialization. Which, to me, is the core ethos of that community.

Somehow, capitalist competition triggers this race toward commercialization. And now all the labs are sort of stuck in their current architecture. It’s about taking the existing transformer architecture and making it bigger, raising more money, versus exploring a problem space.

Were there places doing that kind of research between Xerox PARC and this AI moment? Google’s moonshot factory comes to mind, but were there other examples?

There’s Bell Labs. The semiconductor came out of Bell Labs, but Bell Labs didn't capture it.

There's IBM — PC came out of IBM, did not capture it. SQL database also came out of IBM, did not capture it. Oracle captured it.

A lot of good papers came from Google Brain. They did not capture it. OpenAI captured it. I think the pattern happens. I don't think anything about this era is unique.

Maybe it’s cyclical. At the beginning, there’s not much to protect, so there’s more openness. Now labs like DeepMind, OpenAI, and Anthropic don’t publish as much as they would have otherwise.

Because it was commercialized. You can trace this way back to the steam engine. England banned the export of machinery and the emigration of people with the relevant know-how for quite some time. So the technology didn't diffuse — it was stuck in the UK for some time.

How do you think about open versus closed models?

For open source, at least for the current class of models, the biggest problem is the bio side of things. If that’s taken care of, then to me, it’s net good because it diffuses the technology to more people.

Companies with the resources, the compute, or the know-how will get ahead faster. So the Fortune 500 can get ahead, whereas the Fortune 5,000,000 might not be able to. I see our purpose in this era as helping diffuse the technology to the Fortune 5,000,000.

One of the main constraints is access to cheap compute and inference. Open-sourcing models will help with cheap inference. Open source is net good at the moment. Otherwise, power will concentrate with the people who have frontier models.

Let's come back to Notion. You've described getting access to the GPT-4 APIs as a kind of aha moment. It sounds like it enabled things you'd hoped to do for the last decade.

Yeah. And it sort of made something obsolete. Like, why do we have software? We organize information. It helps us do knowledge work. The GPT-4-class machine can do that as well as some humans, and the GPT-5 class can write code — truly democratizing any software creation. So models can do that better.

In some sense, the original mission needed to evolve. Democratizing the software-toolmaking part becomes superseded by the capabilities, but the democratization nature of the original mission is still the same. Like, how can we be the IKEA or Uniqlo — bringing something good and high-quality to as many people as possible?

I framed it as an opportunity, but there’s also the question of how much of Notion’s original mission can now be accomplished by an outside API. How did you hold both of those thoughts in your head?

It wasn't super clear, actually, before Claude, that you could do really good coding. It was somewhat clear that you could do more and more knowledge work. So what we were thinking was: how do we leverage this to build better tools for knowledge work — tools for thought?

Maybe two years ago, coding became a somewhat solved problem. The original mission of democratizing software toolmaking was solved by better technology rather than through a product. So how do we reconcile this?

It's more like evolution. We had to evolve the company's mission statement from "democratize software toolmaking" to "give people the tools to make their life's work." Now it's broader. It's about providing the tool. Whatever your life's work is, we're going to help you make it happen. That has been the phrasing of the mission statement for the past two years, but the ethos hasn't changed.

I love IKEA’s vision: “To create a better everyday life for the many people.” It’s similar to Charles and Ray Eames’s idea: “the best for the most for the least.”

It’s the same kind of ethos toward making tools. You want to make them as accessible and as high-quality as possible. To me, that’s the essence, whether it’s classic software or classic software plus AI. We’re more flexible about that.

You're a thousand people now, so you can't talk to everybody every single day. There must be chatter in an organization, people having different thoughts when they read things like this in the news. How did you manage the internal dialogue over the last couple of years? How much of it was just doing weekly all-hands and talking about this as much as you could?

Just that. In the early days, people thought AI was another crypto thing. So many people didn’t believe in it.

My interpretation was that maybe people saw it and were afraid of what this would mean for Notion. Instead, you’re saying people weren’t taking it seriously enough.

We were one of the first to move on AI because Simon and I had a huge conviction about it. Not the rest of the industry. Not the people at Notion. Some people quit Notion because they didn’t believe in the AI future. We’re okay with that.

But there was a painful year and a half of grinding where we tried a lot of different things. The first product popped—the writing product. That was fairly easy. It was just out-of-the-box capability.

But we’d been trying to build an agent product ever since that first product, in early 2023. We tried for a year and a half. Nothing worked. Finally, model capability caught up. Now it works.

What were some of the assumptions from late 2022, early 2023 that you've since reversed course on? Some of the experiments didn't work, but what about fundamental beliefs about where this might go?

I think we’re more right than wrong. We just have to execute. I think the learning for us is that we should not jump to the final step. Simon is pretty good at seeing two or three steps ahead, and we’d sort of jump. We learned that you can’t jump—you need to build up to be able to get there. This is the hard lesson from the last two or three years of building with AI.

Can you give me an example of that? What would have been an example of jumping to the end?

We were trying to build an agent product that could use computers in 2023, 2024. But you should build up the capabilities incrementally, starting with tool use and question and answer, then adding computer use as another tool. We didn’t do that. We went straight to the end. It sort of worked, but only 70% of the time. It wasn’t ready for production.

We have a new leader, Max, who joined Notion last year. Simon can see two or three steps ahead. Max can too, but he’s more incremental. There’s a good push and pull between them. So the company’s in a good place right now.

Since the founding of the company, you’ve had different directions you could try. Some ideas are further into the future than the world is ready for. Some are just wrong, or the wrong form factor. But you seem willing to scrap the code and move on. How much of your codebase, over time, has just not been used or been killed?

We’re in the middle of doing this again. We have a project called Apple Cart. Once you tip over the apple cart, the apples roll down and you can’t bring them back.

We called the Japan codebase Notion Next. Everybody in engineering at Notion today still runs on Notion Next. We’ve done this multiple times now.

Our agent harness needs to be rewritten every three or four months. Simon’s really good at this. Just burn it. There’s no sunk cost.

That makes sense from the top-down perspective. But if I’m the manager of a product, it might feel pretty crappy to have my product killed. How do you culturally get people on board with this?

I think if you’re not on board, you’re probably not at Notion.

Okay. So do you celebrate it, for example, at an all-hands? “We killed this and we reset so fast”?

Yeah. Last week, I celebrated a team that unshipped a feature they’d worked on that didn’t work out.

I think we can all do better at that, but those kinds of values flow from the top. Simon's like that. I'm like that. There are no sacred cows. Everything's burnable.

That makes sense. So there’s a willingness to celebrate internal reorientation—reorienting the product strategy and refining how you phrase it.

In most software companies, it would be really hard to make this AI transition. Most musicians and artists stick to one genre. Only the greats can do multiple genres.

Picasso can do two or three genres. There are eras of Picasso. David Bowie has the Berlin era, the Let’s Dance era. Only the good ones have two or three. Everybody else only has one. To me, I look up to David Bowie.

Internally, you can talk about these changes at a weekly all-hands. Externally, you have fewer interactions and fewer shots to put your message out. If one of those messages changes or turns out to be wrong, there's more to claw back from. How do you think about that when you're so willing to move fast?

It's so tricky. Some people's interpretation of Notion is still stuck in the first version—version 1.0, 2.0. They haven't tried our AI product, even though we have one of the best applied AI product teams in the world.

I don't have a good solution. It's as much a craft to change your internal culture and product as it is to change the external impression of you. Maybe it's easier to steal good ideas from the fashion industry, or other industries that are good at reinventing themselves.

So, we should think about that. I haven't thought too much about it yet.

You’ve said there’s a tension between your artistic side and your business side. Did you feel that on day one of the business? Is that feeling bigger or smaller than when you first started?

Mostly art on day one of the business. I didn't need to manage people. I didn't have customer expectations. But then you realize that if it's mostly art, you don't deserve to exist as a business.

Today, if you worry too much about the business, you're building a commodity that is becoming another software or SaaS product. The balance is in the middle. Enough uniqueness, but you want to be accepted by the mass.

Artists do this pretty well. I think it was Andy Warhol who said art is whatever you can get away with. You're on the edge of the thing that you can still direct the masses with. If you're too far from the edge, you're seen as a crazy person. If you're too far within the edge, you're not pushing the boundary.

Would the artist comparison be marketing their work, putting it in galleries, or thinking about the price they’re selling it at? Is that what you mean by an artist having enough of the business side?

Yeah. Are you speaking the vocabulary of the thing that people care about? You can write gibberish when nobody reads it. That is not poetry.

Early on, it was more art because there wasn't a business yet. Today, the business side has grown. If you fast-forward another fifteen years, and Notion is bigger and more valuable, does the business side keep increasing? How do you find the right balance?

I think that’s when those generational businesses are created. They find their place in the universe. They find this equilibrium where what they stand for and what they offer has a sweet spot with human nature, human needs. Not every company can make it.

There's a Japanese concept, Ikigai — the people care about it, the market cares about it, and you can make a living selling that thing. The intersection of this balance for multiple decades, or even for a decade, is how to make a generational institution.

Do you think Notion has found that equilibrium yet?

I would say we've done a better job than other companies in the sense that we're conscious of what we want in the world, our service. But our market has also been changing so dramatically.

If AI doesn't happen, we're going to be stuck in the boring SaaS world for another decade. That's very easy. AI makes the work super exciting, super fun.

It's like the Brazilian racecar driver Senna: if it doesn't rain, he won't be able to take over other people. If it rains, he has a chance at winning. It's raining now, and the field is wide open.

I love that. It also says something about how you see yourselves. You're still the underdog, and there are bigger companies for you to race ahead of.

You've talked a lot about wanting Notion to feel like Legos — highly modular — versus a rigid piece of software designed more for go-to-market. Is that changing with the new AI direction, or is it just taking a different form?

Because AI allows for vibe coding a lot of software, the constraint is no longer your ability to be flexible. The constraint is the balance between flexibility and understandability.

There is a language of how we communicate. There are buttons, forms, text boxes, scrollers. That thing is not going to change super quickly, just like language doesn't change that fast.

It's really easy to create boxes on a screen, but you don't want to create random boxes. For knowledge workers pushing information around their company, you want it to be good enough. You want it to be understandable. So the constraint here is: how do you make things understandable for a large group of people and flexible enough to cater to your business needs? To find that balance — it's what we do.

You said that at one point, Notion had fallen into the trap of making software more like “hard plastic,” trying to design something for your go-to-market organization. What was the hard-plastic version of the product?

Notion is Legos. We give people the primitives, the computing primitives: text, relational databases. With those things, you can kind of have most software. A project tracker is kind of like a relational database of tasks and projects, with a relation connecting them.

The advantage of Notion's approach is that it's very flexible, but it's not as smooth as a product dedicated specifically to task tracking. Products like the Atlassian suite or Linear are dedicated to that. So they can make a product that’s really good for that use case out of the box, but hard to change. We're on the other end of the spectrum.

To compete with that head-on, one easy route is: with Legos, you can have the hard-plastic version of a horse, for example. You could ship that version of the horse to compete with Atlassian's horses.

We sort of caved and tried that for a year. It didn't quite work internally because of our culture. People just didn't feel comfortable doing that. Our codebase wasn't designed for that. There were a lot of reasons. It was like internal organ rejection.

Our customer community was kind of confused. The commitment they have from Notion, the Notion they know, is, “They provide Legos. We build our own thing.” So it was like, “Why are you guys shipping this horse piece?”

That was sort of an aha moment for me, three years ago, around the beginning of the AI era, actually. We have to stay true to our values. That goes back to your point of how do you build something generational. Your ethos has a large enough clientele in the market that wants what you provide, and you're good about providing that, and you can do that for a decade. If you break that ethos, you become somebody else. And nobody asks for that.

Your design aesthetic shows up in the product, but also in how you dress and in the office. There are historic moments in technology and well-designed objects framed outside the rooms. There are also Muppets in the conference rooms. Tell me about those examples. Why did they emerge?

We want some token to represent a customer in every room. And Notion is kind of like Lego building blocks, so we want to be playful. I never grew up watching Sesame Street, but they make sense. So every room has one or two Sesame Street characters.

We named the conference rooms after timeless tools. We want to be inspired to create timeless things. Software is not that timeless today.

Do you still practice some of those childhood hobbies, like watercolor or painting? Or does the creative side all come out here now?

My creative output today? I haven’t designed directly for a few years, though I got back into coding recently with coding agents. I think the creative output for me feels like giving feedback, growing the team and people. This is my clay. This is the thing I’ve been shaping. Shaping the culture.

As I've gotten older, I'm realizing I'm more of a journey person than a destination person. And a journey person is more about who are you on the journey with? And what’s roughly the purpose of being on this journey? And you just want to enjoy your time surrounded by good people and make sure the journey is meaningful.

I respect that a lot. I feel really similarly. If we fast-forward, there will be a day when Ivan’s not the CEO of Notion. God willing, it’s around for another hundred years. How do you think about going from a founder who created this place and its culture to what eventually will be handed off to somebody else?

I think going back to the earlier point we talked about, a company that is durable has integrity. It's kinda like the thing can fight off entropy on its own because it has integrity. It attracts people. Customers want it. The equilibrium's gotten reached.

In some sense, the founder CEO's job is to fight off entropy trying to harm its integrity. If the integrity is pretty strong, it will live beyond the people who started it.

Steve Jobs hasn't been at Apple for over a decade. Yes, you can argue it's not like the Apple before, but he did an excellent job, so it still resembles the Apple of the past more than not.

When you think about Notion’s flexible, Lego-like approach to building products, what companies have you taken inspiration from, inside or outside software?

Well, the earlier computing companies we mentioned. But also Lego itself. Think about it. That company lasts kinda forever. As long as we have hands, that company is good. It almost went bankrupt in the early 2000s, but it came back.

IKEA is another one. The founder started in the 1940s. Then he ran it as a small business. In the 1970s, he put out his company mission statement. You should look it up. It's called The Testament of a Furniture Dealer. They still live up to that today.

You talk about the long-term journey for Notion, surviving through AI and whatever comes next. Is part of that your approach to the finances of the business? You haven’t made yourself dependent on venture capital financing. You’re a break-even and profitable business.

Pretty early on. I think we started doing this kind of small-business, small-team thing before everybody else did. We made a lot of mistakes during COVID, but one thing I appreciate is that we didn’t blow up the company in terms of headcount. We grew a lot, but other companies were expanding their teams much more aggressively.

I think that comes back to Simon and me. The two of us have a lot of skill sets. Between us, we can run the whole company. Marketing, product, sales, coding, design. We want to build a company in our image.

You're in control of your own destiny, and you don't have to worry about someone else's timeline. It's your timeline and your journey.

Yeah. That's after the fact. But we're lucky to be surrounded by investors who've been with us for a long time. Patient.

There's Ivan the person, and then there's Ivan the CEO and founder of Notion. What's the end-of-career, end-of-life North Star for both of those people?

I can look back and say, hey: I did something meaningful with the group of people I like for a long period of time, and the meaningful thing helped other people in some meaningful ways. That's the overlap.