Notion CEO Ivan Zhao on if AI is really killing software companies

Support for today's show comes from Anthropic, the team behind Claude. You know how sometimes a problem just grabs you like you sit down thinking it's a quick thing and suddenly it's midnight. That's exactly the kind of mind Claude is built for. People who don't just want the answer, they want to chase the thing that's underneath it. Anthropic positions Claude as a thinking partner, not a search engine. It works through the problem with you and it doesn't try to wrap things up neatly, which if you've spent any time on this show, you know that's a feature, not a bug. Get started with Claude for free at Claude AI Access and see why problem solvers choose Claude as their thinking partner. Support for this show comes from Vanta. Vanta uses AI and automation to get you compliant fast, simplify your audit process and unblock deals so you can prove to customers that you take security seriously. You can think of Vanta as your always on AI powered security expert who's scales with you. That's why top startups like Cursor, Linear and Replit use Vanta to get and stay secure. Get started@vanta.com Vox that's V A N T A.com Vox vanta.com Vox.

0:00

Definitely we're looking for fair AI native people. We're shifting towards younger early career folks.

1:17

Early career? You hired a 16 year old the other day.

1:22

Wait, did you really?

1:26

Yeah. Tell me about this. How do you know about it?

1:27

Welcome to Access and hello to all of our new listeners. Viewers, after last week's splashy episode with Fiji Simo reminder, find us on YouTube in beautiful 1080p or higher definition. Ellis, how are you doing my friend?

1:32

I am something. I am tired of over committing my calendar, my friend. As a new solopreneur who is also deeply a people pleaser, I set myself up today with an hour of therapy, an hour of client review, 45 minutes for lunch, a two hour work session with a client, and one hour podcast where I had to bring it and then now we're doing this. So yeah, my lips are chapped, a little worse for wear, but I'm all right.

1:49

Sorry, 45 minutes for lunch? I don't think I've given more than five minutes for lunch on a weekday in 10 years.

2:21

Well, thanks, that makes me feel better.

2:27

I'm just saying man, like you had some pad time.

2:30

What kind of co host are you? You gotta validate, support and praise the co host. I mean, if I could have an agent manage my calendar and be like, ellis, it's just too much. Then it wouldn't. At least I wouldn't have to be the one to say no, say, my agent told me no.

2:33

Well, we can ask Ivan Zhao, the CEO of Notion, who is our guest today, about that in a little bit. But first, Ellis, I want to hear about your time up in San Francisco last week. I'm usually the one who's going to San Francisco doing the JSX up from Burbank, but here you were last week with me, staying put in la, going up there and having some fun and telling the founders what's up.

2:51

Yeah, it was a lot of fun. I was invited by their speedrun accelerator, Andreessen Horowitz, A16Z, to speak to their brand new cohort of young founders about the art and science of startup storytelling. Of course. And I think it's gonna get posted online later. And it was a bit of exclusive for now, but I can share the things that people tweeted about it and then I could say more. I think the big message, which I came up with the morning before because I was trying to do my storytelling method on myself and say what people actually want to hear. The big message was don't say what it is, meaning don't say what your product is, but say what people want. And I tried to give a bunch of examples of kind of how I and others have done that through history. I think in a world where tech is kind of, frankly a lot less differentiated as it once was, where everybody had different innovations, now everybody's kind of talking about the same AI stuff, as powerful as it is, talking about agents or AI or saving time and money is going to be tough for you to stand out. Right. And so I think I just kind of shared the inside scoop on my method of trying to find, putting my reporter hat back on, trying to find whatever's most interesting about your company through the eyes of your audience and just like hammering that home, you know? And I think like one of the examples that resonated with people was when I worked on the Daylight tablet computer. And I think being tech dudes who did have like a legit innovation, like a 60 frames per second, super responsive scrollable E Ink touchscreen so you could browse the web, you could listen to music or podcasts if you want, you could read your newsletters inside your email. It's kind of like a Kindle that can do a lot more. Something like between the Kindle and the iPad. I think any other tech company would have just talked about the screen technology, especially if it was a company like Samsung, but the fact that they also created their own os, that was all kind of about calm and peace and focus. And it doesn't want to have notifications, it doesn't want to have social apps. They try and move you toward a psychological knowledge workers Zen moment. We ended up making the headline about the larger idea, which is, I think what I believe people want most is just a computer that's calmer and more focused. And so the headline I came up with was the computer de invented. When everybody else is reinventing and adding more and more and more into their products, I think reinvention has become synonymous with just adding more inshidified stuff into your product. What if there was a de invented way? And as always, the headline is just a hook, a way to get people in. I don't know if you'd like it, Alex, but.

3:15

Well, I would just say that you weren't very calm when you brought your daylight to our live show with Google Ventures and Winston from Harvey, and that shit broke immediately and you had no notes. But yeah, if it works. I can see that working.

6:03

Yeah. Android is not my favorite to this day. I mean, I think half the problem was that I couldn't make the text bigger, coincidentally, in the Notion app. I know where that setting lives in iOS. I do not know where the setting lives in Android. Actually, I think I tried to change it, but it wasn't reflected in specific apps.

6:15

So, yeah, it's okay. I want to know about the actual day of doing this keynote, like, and the process of it. Were you. Were you pacing? You know, were you doing like, you know, you go to one side of the stage and then the other. Did you have a teleprompter? Were you winging it? Do you have a bunch of slides?

6:33

Slides, but a lot of riffing, which I am prone to do as a yapper. That's why I have a podcast. I was supposed to go 45 minutes, but I went 20 minutes over.

6:50

Jesus Christ. You talked for over an hour straight.

7:04

Yeah, I mean, you heard. I've talked for like seven hours today already, including this. So, yeah, I guess I can do that, but I just get excited. I mean, I like to make people laugh. I think of myself as a little bit of an entertainer or a standup comic. It's interesting though. I felt like when I got out of media, I was like, I don't want to be in media because I don't want to be an entertainer. But I do think going back to the People pleaser mindset. That is definitely a part of me.

7:07

Yeah. Did any 20 year old founders come up to you after and be like, I was five years old when you

7:34

were at the Verge? Shockingly, no. They all came up to me, which was fun. I mean, they are founders, meaning they are ultra networkers, ultra ambitious. They, by definition, are the people who approach the speaker after the event. But in a room full of founders, all of them approached the speaker after the event, which is funny, but, yeah, I don't think any of them were alive when I was at the Verge.

7:41

Isn't that wild? Something else you did last week after we recorded was you had another company that you worked with, Go Live Faces. Is that right? Yeah, this looked interesting. Like a new kind of PowerPoint. There's a million companies doing this. But I liked the kind of weirdly irl, low production look of at least the still of this trailer. You know, just chairs on a beach.

8:03

Yeah, this one was fun. Essentially what they're doing is instead of having your slides just be these kind of static rectangles, they are using the full power of the web to create live interactive experiences inside of your slides. Whether it is a rotating 3D model, a live map, a quote you can generate from a client. Anything that can be on your site can now be in your deck. And this was kind of like the perfect client for me because they have this very, very cool tech, but they aren't completely sure where they want to point it. And so for me, the exercise is always about, what do you love? And they talked about founders building, cool stuff. What's the technology you have? Where might people want it? Who do you want to compete against? And I think we ended up in a really cool place. This was also a line I was proud of from the manifesto that I also tweeted. There's no power left in PowerPoint that, you know, you just got to give the people what they want. Right. Everybody loves that shit.

8:30

So again, what were you showering when you did that one?

9:28

Dude, I, like, I, like, don't. I don't even know at this point whose idea it was. Could have been me, could have been Cloud.

9:32

You're so in the zone.

9:40

Yeah, I guess so. I don't know, I feel like I'm burning out a little bit in terms of my, like, random access memory. Like, I don't even know what happened yesterday, but whatever I'm doing seems to be working at the moment.

9:41

Well, we have to talk about the biggest news of this week, which is that the founder of OpenClaw. Peter Steinberger is joining OpenAI. So you, you haven't, you haven't set up open call, right, Ellis?

9:56

Oh my God. Did anybody make the joke that Altman buttered him up?

10:09

Oh, what, because he's a lobster?

10:15

Exactly.

10:17

No, because it's a bad.

10:18

Not my best.

10:19

Not your best? That's okay. If we were in person, I would laugh.

10:20

But there's some legendary buttering up methodologies, are there not? Zuck used to take people for a walk around campus up to that hill or whatever. I don't know. What do you think Sam does?

10:23

That's a good question. You know, I asked someone very close to this conversation why Peter probably picked OpenAI and Sam Altman over Meta and Mark Zuckerberg, because he had actually said on Lex Friedman's podcast that those were the two offers he was evaluating, which I just love how the entire process has been public and on display for like the last two weeks of. He went to San Francisco, he talked about having these conversations, and then he joined OpenAI, like all within two weeks. But anyway, I was talking to someone very close to this and they said probably money was the reason he went to OpenAI, though. I don't know. I mean, Steinberger is an interesting dude, man. Like, he's, he's a repeat entrepreneur. He sold a company that did PDF stuff for like over $100 million. He did 40 something projects before Claude.

10:33

He's the PDF goat.

11:29

He's the PDF goat. He's also jacked, living in Europe, and it doesn't seem like he really needs a lot. I mean, on Lex Friedman he was talking about that, but I actually just wanted to play you him talking about the offers he was getting. Let's cue this first one.

11:30

And then there's all the big labs that I've been talking to. And from Those, Meta and OpenAI seemed the most interesting.

11:47

Do you lean one way or the other?

11:56

Not sure. Amata should share that. It's not quite finalized yet. Let's just say, like on either of these, my conditions are that the project stays open source. That it maybe it's going to be a model like Chrome and Chromium. This is too important to just give to a company and make it theirs.

11:58

So yeah, this literally came out, I think a couple days before he joined OpenAI.

12:18

Dude, I don't get this at all.

12:22

You don't get it at all.

12:24

It's so fucking simple. One company is about personal productivity and the name Open is in their name and they have 800 million intentional AI users every week one has 800 million unintentional AI users every week has not stood for personal independence and doesn't really have anything to do with. With productivity or owning your own shit.

12:25

Okay, fair.

12:46

Am I wrong?

12:48

That is a very fair point. I think an interesting argument is that Zucker is over there trying to assemble the Avengers of AI to reboot everything and is willing to consider things he hasn't considered before. But yes, your point is well taken. I think it also helps that Open Claw ran on Codex, OpenAI's coding platform and that Steinberger was such a proponent of that. Yeah, point taken. Now the question is, what is he going to do at OpenAI? I'm not so much concerned about the future of Openclaw because again, it's still this super early adopter thing that most people, including myself, will not take the time to go through and spend thousands of tokens and buy a Mac Mini to like have it, maybe triage email. Okay. I would rather just wait for the hosted safe version of this, which, by the way, Kiwi and then Manus, which Metabot shipped like also this week. So OpenAI will do a version of that. But I think there's actually something else Peter will work on that connects to our last episode with Fiji and he was talking about this with Lex Friedman too. Let's play this.

12:50

There should be a very easy way for agents to get their own Twitter account. We need to rethink social platforms a little bit. If we go towards a future where everyone has their agent and agents maybe have their own Instagram profiles or Twitter accounts or can, like, do stuff on my behalf. I think it should very clearly be marked that they are doing stuff on my behalf and it's not me because content is now so cheap eyeballs are the expensive part.

13:58

So, yeah, I think he's going to be working on this OpenAI social network for agents, which Fiji told us about for the first time last week. Let's just quickly play that, you know,

14:30

nothing really specific in the works, but the thing we're thinking a lot about, and I think everyone is thinking a lot about, is like, what happens to, like, social relationship in a world where everybody has their own personal agents and like, how can these personal agents maybe help you manage your social relationship in a better way? Like if your agent and my agent were like, communicating ahead of this interview with that interview have been better.

14:41

So. Yeah. What do you think about this, Alice? It sounds like we're going towards a future where agents are talking to agents.

15:07

Well, he certainly sounds like A PDF goat. I'll put it that way, man. I don't know, it just sounds like one use case to me. I mean, when I think about what I think the Johnny and Sam device is going to be, it is this always on thing that has access to computer use, browser use, app use APIs, MCP, whatever it is, is going to be on the back end and you can use it to orchestrate whatever you want. And that could be an automation, that could be something that you want your computer to do while you're away. I don't know if they're going to use this guy's open source technology or even just the principles of it to help build it. But, but tell me this, I mean, I'm still not completely sure. I'm not completely sure what, what was the hardest part about what he did?

15:14

That's a good question. It sounds like a lot of it was around memory, around the way that he designed it, not so much the underlying technology because he was using, you know, open clause based on models like CLAUDE and GPT and all that. So it's not like he has some proprietary model.

16:03

I think it was connected the dots.

16:20

He connected the dots and let it, and let it go wild. And which is what ChatGPT did originally was, you know, Google and all these other companies had chatbots like that and they just didn't want to release them because they didn't think people were ready, they didn't think it was safe or secure enough. And this open source project is like, here, just have at it. Install this thing on a Mac mini, spend a shit ton in tokens and figure it out. And I think what he showed is that people, at least a significant early adopter crowd, which is probably also listening to a show like this, want to do that even if the ROI on the other side is not known. And so if you take that energy and you scale it to 100 million, 800 million people with the right safeguards, with the right tool access and really abstract away the difficulty of setting it up, maybe you've got something pretty cool.

16:21

Yeah, I think the other thing he hinted at with this project is back toward this vision of the future that there's always been where you own your computer and your computer does things for you versus let's say kind of the approach where it lives on a corporate server and then that stuff does it for you. But I mean, I'll tell you this as a segue into this notion. Interview with Ivan Zhao I mean I was using the new custom agents tool and Basically what I was able to say is every morning at 7am, like run a script where you update my availability document based on when the next two hour slot is in my calendar for big project learning sessions and also the next one hour slot for one hour working sessions. And it's just going to do that every day. And then I made up another one that said whenever I get an email or I said after the final review with a client on my calendar, go to the relevant invoice pre, fill that into an email and get it ready for me.

17:09

And it works.

18:09

Not yet.

18:11

There we go.

18:13

There's still some kinks to work out. One of them is a simpler kink which is that the original availability one couldn't access my sub calendars. That'll get worked out. It's on my main calendar just fine. But I'm a nerd who has multiple sub calendars for each client, so that's my problem, I guess. But as far as I know, being able to like create time based or rule based or like trigger based stuff is a lot of, you know, what people are talking about when it comes to openclaw and whatever is talking about. I think part of it is that you could kind of own it and run it yourself and have it be always on as a result.

18:15

Yeah, but I don't think most people want to do that.

18:53

Oh for sure, for sure.

18:55

Yeah, I think most people want the Manus version, which Metabot, which is like it's hosted for you and you access it via Telegram or WhatsApp and it's like it's your own little cloud that you control, at least theoretically. But yeah, you'll always, you know, it's the same people that have like physical Bitcoin wallets. There's always going to be people who want to control it on Prem.

18:56

And I mean Ivan actually gave us a juicy little hint here on the show about like this independent app that they're working on just kind of talking to the AI and it's going to be able to do stuff just like that.

19:16

There's actually a lot of interesting new stuff in this convo with Ivan. He is a super well connected, influential dude, obviously in AI and tech more broadly, but also Notion I think has a soft spot in both of our hearts. You worked with Ivan on some of the company's positioning at one point. Just full disclosure, like we're friendly with Ivan. Notion was actually one of our launch sponsors for Access, so mad props to them for that and just generally really good people. So it was super exciting to finally get time with Ivan on the show.

19:28

Do you want to hear any more from Delirious seven Hours Talking Ellis, or should we just cut to Ivan?

19:59

Let's just cut to Ivan.

20:05

Okay, have fun. Support for this episode comes from Anthropic, the team behind Claude. I've been using Claude to help me resolve those messy contradictions at the heart of the brand. Strategy how do you triangulate one message across a company's products, beliefs and future vision? Claude doesn't just tell me what I want to hear. It feeds me different strategies across a spectrum of approaches. More like a partner, less like an answering machine. When you need to get up to speed fast on something complicated, you can just point Claude at it and it pulls together sources, surfaces where people disagree, flags the things you should be asking about, all with citations. You can go check when you connect Claude's Cowork feature to your Gmail and Google Drive, Claude will read your actual emails and docs along with you instead of waiting for you to paste things in. Then Cowork takes it further. You can point it at a folder, describe what you need done, and Claude just handles it. Queue something up before lunch and come back to finished work. Try it yourself at Claude AI Access

20:06

it's true that some things change as we get older. But if you're a woman over 40 and you're dealing with insomnia, brain fog, moodiness and weight gain, you don't have to accept it as just another part of aging. And with MITI Health, you can get help and stop pushing through it alone. The experts at MITI understand that all these symptoms can be connected to the hormonal changes that happen around menopause. And MITI can help you feel more like yourself again. Many healthcare providers aren't trained to treat or even recognize menopause symptoms. MITI clinicians are menopause experts. They're dedicated to providing safe, effective, FDA approved solutions for dozens of hormonal symptoms, not just hot flashes. Most importantly, they're covered by insurance. 91% of MITI patients get relief from symptoms within just two months. You deserve to feel great. Book your virtual Visit today@joinmidi.com that's joinmidi.com.

21:08

Ivan, how are you?

22:13

I'm pretty good. How are you guys?

22:15

Good. Where are you calling in from today?

22:17

I'm calling from our San Francisco office.

22:20

Are you tired of hearing and saying the word agents yet, Ivan? Are you still loving it?

22:23

You know one of our best designers in the company. Do you guys know this guy Jeffrey?

22:29

I don't think so. Sounds like I need to meet him.

22:36

Yeah, he's great. But whenever he says Asian, Asian sounds like Asians, we have this. Asians do this and he's Asian. Right. So it's like Asian, Asian. Okay, it makes sense too.

22:39

So we're catching you at an interesting time week before the big rollout. I know you previewed custom agents a little while ago, but how's everything going toward the launch?

22:53

That's very good. We're just prepping up for all the custom agents ready to launch next week. Been testing for a while. We've been actually talking about today in all hands. Like we've been working on this idea almost a year now. Next week will be the week that ready finally ready for public. So we previewed this earlier late last year in our MakequinOcean keynote events. So bunch of company will be running out for some time. But next week is the public time.

23:02

Ivan. It's funny, I feel like I was giving custom agents a try. Big thanks Taksha for the demo with Alex the other day. And what I was thinking was that it was kind of like your initial vision years ago for being able to create your own software. But you guys dumped that line, right? Is it time to bring it back? Because it really feels like that democratized

23:30

software right now is democratized AI agent. It's kind of similar. People are trying it from all different angles. You can even see from what is it openclaw in the past couple of weeks, several weeks blowing up, right? People buy Mac Mini they want to use. Having asynchronous agents that are running on the background can do work. But what is like to have a business grade multiplayer. Then you can create a factory of those kind of customer agents working 247 for you and open up easy enough that more people can do that. That's the spirit of the product we're launching.

23:54

Ivan, have you actually had openclaw do anything good for you? I'm actually dying to know anyone who's had openclaw do something meaningful for them.

24:30

One of our head of product. I don't know if you hook up opencloud but a lot of cloud code to manage his lights for him. He has those Lutron systems. He figured out how to plug into his AI system whether it's open cloud based or his only agent system.

24:40

Maybe not worth thousands of tokens in a Mac Mini running on its own, but we'll get there.

25:00

I think we all have to tinker within us, right? We all want to. People play Lego part of it to see the final result. But I would say the journey is actually making the Lego pieces. And when you make something great, you want to showcase to people. I think it's a lot of the draw for it, including Notion included. People want to build something that showcase to other people. So Open Cloud plays into that too.

25:07

It's tough, I feel like with any of these big paradigm shifts. And you're uniquely good at this, I think Ivan coming up with like historical references or metaphors. And for a while I feel like it was Legos. Now we're moving on to the factory. And I feel like I had a founder client who was talking about that game Factorio. And I have a special thing that I want to show you from my. From my history that I just thought of that's actually right behind me. Hold on one second. Uh oh, I have the original PC box, one of the best box arts of all time, SimCity 2000. And like, it does feel like that. Like you're trying to manage and optimize this incredible system all by yourself. But what's crazy is that you could just. You're. You almost feel like the boss. You feel like the CEO, like the mayor of your Sim City when you just say it with words now. And that's. To me, I think what's, what's the craziest piece with this empire you've built that is now accessible with the agents.

25:28

It's actually. So there's two folds. One is, that's the product we built. We want all the leaders, CEOs, whoever working company feel like they can manage a team of agents or interns.

26:22

I can't unhear it now, Ivan.

26:37

That's a product built. But at the same time, like me running Notions, I've been doing a lot of with coding agents just plugging all the different tools inside the company. It just feels like superpower. Like some of our customer agents connect with Snowflake on one end. All the context in Notion cloud code can run some other process on the other end. Like I can ask any question inside the company and the answer will be it's a superhuman level of answer for me. And I can run some repetitive processes and I can almost feel like I can do anything. So that power is, I would say until this coding agent revolution with since 4.6, we haven't have that power. And that's fairly new. It feels like Sim City, but for your companies.

26:41

Akshay was showing us how you guys use the custom agents internally at Notion, and I thought that was pretty interesting. Like Everyone you guys run on Slack. And he was showing us how it can monitor one of the channels where people are asking like basic things around the office, like where are the chargers? You know, I'm out of whatever at my desk. And it can respond and remind people of inventory and if not create a ticket that a human can then go and check on. And I thought that was a pretty cool use case because it's not, you know, revolutionary, but it does lessen the effort that you need to get basic things done that humans would have done before. So I'd be curious to see how that scales out and what other people build.

27:27

Yeah, it's one of the. So there are three use cases we try to really nail with this custom agent launch. So you just mentioned one of them which is answer repetitive questions, a big swap of knowledge. Or you can argue maybe a quarter of them are just answering questions that live in some kind of knowledge base. So we try to do a really good job of that. Another one's triage tickets, triage to dos. You probably saw some demos from hey, people ask questions in one Slack channel, you should pass the team in the other slack channel. A lot of PM's job, a lot of engineer manage job is doing that. The third one is kind of write this kind of status report, status updates. So many teamwork is like hey, what did we do last week? What did you do yesterday? So you don't need to do those anymore. And those three type of use cases, what we consider busy work and most knowledge works actually a big swap of knowledge or are those kind of busy work?

28:13

It sucks. It's so much tedious stuff, man.

29:08

I know agents can do them pretty well in some sense. Superhuman level. Well now so just delegate those to agents and you human can elevate up and play your sim cities, right?

29:11

I get to finally be the systems thinker I think that I am when you're forced to actually construct the perfect instruction manual for each agent. And that was, I think one of the things that you guys have done really well with this launch is create that settings panel where you can actually fine tune the different prompts, the different access, the different integrations. And I've been playing a lot with Wabi recently to create some little vibe code apps and they've done something really similar and I think we're seeing a very cool innovation. People are moving beyond just the chat screen to actually catalog some of its memories in a UX that actually makes sense. Like inside of Wabi there's a settings page within your app. For the icon prompt, there's a setting page. For the different fetch prompts it's doing and you can select a model and I feel like you guys have taken a similar step forward. When did that settings page come about? To let people actually tweak things not just with natural language, but in more of an interface. It feels like a big. A big next step for AI.

29:23

I wouldn't say the big next step is about required step to do anything business facing because when you're working with a lot of people just like giving

30:25

the transparency into what it's actually doing

30:33

a transparency and you want to lock down, you want to make sure the permission is set. Like when you work with a team of people, there are secrets. Certain people should see certain things, certain people should not see certain things. A certain workflow should be defined by the experts and people shouldn't change it. The power of language model is free for all, anybody. If you can speak English or Japanese, you can modify things. The downside of language model is free for all. Anybody can modify anything in a business setting. You don't want free for all. You want some guardrails for the permissions, for the securities, for all the boring but very important and necessary things. And in some sense that's a lot of value we provide with customer agent to provide those guardrails. So like open cloud, you can run a Mac Mini to run everything your life, but that's just by yourself. But what is that for a company in that unique guardrails and all those checkboxes you mentioned?

30:35

Yeah, I'm curious. I would think you guys are on the cutting edge of actually implementing this stuff internally. I feel like every single company out there is saying like AI agents are going to help you automate the busy work and focus on what matters. But what have you heard from the team for what that actually looks like in terms of being able to focus on what matters?

31:28

It's the repetitive work we're talking about. We've been testing this with a small group of alpha customers like the Ramp, the Vercel, the Clay, the Cursor of the World, the Access podcast.

31:49

We've got early access.

32:02

Yes. Even though those companies are very AI first, they have a lot of busy work too. They can wipe code, a lot of internal workflow on a coding side, but anything that's knowledge, work related, teamwork related notion is the best primitive for that. Give you one example, Ramp fully onboarding to notion last year. Right now I think it's about 3000 ish people. Companies before notion they have to use across like half a dozen core collaboration tool. So they just shrink from six just to notion. That's the first step, consolidation. So save costs, save money, move faster. One place for information. But now once you have one place for information is have one context for AI agent. So they deploy notion AI across all the ramp employees in one place for AI meeting nodes. Now there has one place to do the custom agents so they can actually automate some of the repetitive workflow inside Ramp. One of their popular ones, it's for their sales teams. So ramp is shipping really fast. They're shipping feature left and right and it's really hard for their sales team to keep up. Right. Like what's the latest feature? What's how do you answer question from customers? So now they set up a customer agent that can answer all those questions from all the slack channels, internal notes and docs and emails. I think that's the most popular one inside the company. In the last couple of weeks they I believe answered 4,000 questions for this customer agent. And that's about, if we consider each question is about 30 minutes, that's about 2,000 human hours. That's a real time saving for Ramp.

32:03

It's funny when you say 2000 human hours, I think CEOs love that kind of talk, but I feel like the AI narrative could have been a little different had it been a bit more targeted around the exact use cases. If you were to say hey help desk, you don't have to tell people where the bullshit is that they're asking for every day. Like the IT help desk, they would have been like oh my God, give me this thing. And that's like one of the reasons that in my work, as you've seen Ivan, I always try and lead with like the hyper specific relatable problems because it instantly makes this stuff feel a bit less threatening. But then the question remains, like with all that stuff automated, what are people doing instead? Like have you seen the pace increase for your own team now that they're not doing the busy work? Like what's the outcomes that you're actually seeing? So I haven't heard this from anybody yet.

33:37

Yeah, I will say the first appearance is happening on a lot of engineer side engineer product design within notion. Like we don't all the daily stand up inside notion were done by customer agent. Now so you don't have to summarize the meeting notes, you don't have to say what you did last week. Right. Those are small time saving like a 5, 10 minutes each meeting. But they Add up other things like triaging tickets, managing the bug backlog, none of this done by any PM EM anymore. They're still small. There are like 5, 10 minutes here and there, but they add up. I will say the shape of the time saving is very horizontal. It's not very vertical. Meaning that it's 5 minutes, 10 minutes here and there for each everyone versus the entire role or entire job done by agents. Because of that, we don't see a structural change, at least in company yet. It's like everybody's life gets a little bit better. If you ask our team, like, hey, we're going to take away the customization from you and we're going to have a revolution inside a company. Right. But it does not precisely map to any persons responsibly at this moment because the model capability to get better, the guardrails and all the permission stuff need to get better.

34:25

That's what we learned with openclaw, right? Is that it's cool to let it go wild, but the model is still kind of unreliable. The safety is just not there. There's prompt injection. There's all these reasons you don't really want to let some agent go ham with shell access to your machine and I guess that's the opportunity notion is is going after. Is that fair that this is like fair?

35:34

Yeah, it's like what is open cloud spirit asynchronous can do almost everything you can do with a computer, but make a multiplayer make a collaborative with enterprise grade permission and security.

35:57

How do you use it? Like you talked about how some of your team's using it, but you're the CEO, like how are you using agents in your day to day to run

36:10

the Every morning I have a brief from this guy. So this is out of the box. So it tells me. Actually let me read my brief today.

36:18

So if you're going to talk to these two schmucks on a podcast, I

36:27

was going to say please cancel all meetings with people that are five skip levels below me that somehow got on my calendar. Thank you, bye.

36:29

Give me a daily brief every morning. I used to archive my inbox. So we also notion mail notion calendar team. Their top priority right now is to make their agent to use capability world class, the best in class. So you can archive any. I basically don't check my inbox like a normal person anymore. I largely talk to the customer agent that runs my inbox. So it has a context of what I'm working on, has a context of what kind of thing I care about. Right so that's another thing, the super power one I mentioned earlier. Connect to our Snowflake backend so I can understand almost any data inside the company with the commentary. That's the contact from Slack, contact from Notion Docs from an email. It's kind of like a data science team in your pocket.

36:37

What was your brief like this morning?

37:24

I mean, I usually once you dismiss it, it goes away. So let me.

37:26

That's kind of crazy that you're not really looking at email anymore. I love that. I would love to not look at email anymore.

37:31

Happy to set up for you guys because it really changed your.

37:36

All right, so we'll stay on after the pod, Ivan, and you'll personally walk us through.

37:40

Right.

37:45

I was just doing this yesterday with. Who's that lady from Wall Street Journal. He said he finally clicked with Notion.

37:47

Oh, is this why Joanna tweeted that?

37:54

Yes. Yes. Yeah.

37:56

Oh, Joanna. Shout out. Joanna. She just left.

37:57

I was going to bring up her tweet. She was like, I finally understand Notion thanks to AI.

38:00

And I was like, that was you, Ivan. You made her understand AI, Ivan.

38:04

That's the power of going direct, as they say.

38:08

It says a lot about our onboarding. Could be better. But it also says our AI is quite magical. Right. Can do everything you asked for. So I actually DM her afterwards. We're going to set up one on one to truly onboard.

38:11

That is the thing about Notion for me is like, it still is kind of like, man, there's so much to do in there and I'm not sure where to start. The onboarding has to be tricky when it's such a powerful tool.

38:23

Yeah, I would say in some cases that has been our weaknesses for years because we give people the Legos.

38:35

Right.

38:41

And this moment seems to be a blessing because if you think about traditional SaaS, you hard code a workflow into your software logic and it's like that's your domain, what you're selling. Everything's rigid, but it's powerful, it's automated, it's on the cloud.

38:43

Some Salesforce dude spends eight weeks programming it for you.

38:58

Well, Salesforce is actually a pretty flexible one compared to many vertical SaaS. So actually as a relational database behind the scene, you can do a lot of different things. But majority of SaaS is pretty rigid, the workflows baked into it. Notion's Legos would try not to be very opinionated. So historically it's difficult to onboard, but with language models, language model can use those Legos. So it's actually make our weakness into Our strength. So now you can do all the different open ended things. All different flavor knowledge were in notion so it actually shifted a little bit.

39:02

I have personally onboarded Alex to a variety of different use cases and I will require a small commission. Thank you, Ivan. So, you know it's funny, my barrier to getting on notion back when I joined the browser company is and I actually hated it for a hot minute now I'm like an evangelist essentially. But I was like where the fuck do I click? There's six buttons over here, there's six little dots over here, there's a little table dragger over here. You could drag around the lines. And now that I know how to use it, it's like my favorite thing. And I use Google Docs and I frankly want to kill myself. But there are all those little things I think that kind of add up. And another really big thing I think to learn is how to actually use databases well which was such a huge unlock for me. Are you telling custom agents to proactively recommend stuff like that that maybe is underappreciate? People don't even know what they could ask for yet. They ask for a faster horse instead of a database.

39:36

We don't proactively ask agent to suggest a workflow yet. We're actually testing this when we're deploying customers. It's like how do you automate for deployment motion? This is a buzzword forward deployment engineer. How do you make that better? With our own tools we should do that. I would say the most popular use case is rather than you build a database from scratch like the rows and column and relational schemas, Agent can just build it for you. And that has been at this point it's more than 50% of databases are built by agent than human. Now that's probably the magical thing that clicked with the Wall Street Journal lady yesterday I found my daily reminder daily brief. So my top three items is confirm our domain because we're trying to. I think we're trying to buy another domain for our product.

40:32

And number two, don't let that out here until because it's coming out Thursday, you got to get the domain before

41:20

then the other one's prepare for podcast block this afternoon with you guys. There's a bunch of links in my calendar email and the last one before end of day, review the candidates in my recruiting databases. So we have a recruiting database internally. Every day our customer agents send me a notification. Those are the people who are job offers going out. So I usually review those job offers.

41:26

Speaking of Recruiting. I know some CEOs like Toby Lutke of Shopify have been pretty forward about, you know, they're essentially saying you are not allowed to hire anybody unless you could prove that I can't do it. Whereas other companies have said whoever we hire has to be provably AI native in all their, in all their workflows. Are you implementing anything like that yet in the recruiting process?

41:49

As criteria, not strict, but definitely we're looking for very AI native people. I would say our overall we're shifting towards younger, early career folks because they're just.

42:13

Early career. You hired a 16 year old the other day.

42:24

Wait, did you really?

42:28

Yeah. Tell me about this. How do you know about it?

42:29

I saw a tweet, a 16 and 18 year old.

42:31

Okay, so the story is late last year we hired, well may last year we hired at 18 years old. We thought this is the youngest what we ever hire. Right. So still just graduated high school, going to college. Then a month later we found this person really talented on YouTube, not famous on Twitter anywhere. He making some cool videos about design, AI stuff. So come, we invite this person to our office. Then I bump into Tim in the elevator, say, hey, where do you go to college? Say, I don't go to college. So you drop out, right? So no, I'm still in grade 10 in high school. But this kid is so talented and he's actually part of our.

42:34

He runs marketing.

43:19

No, no, no, he's a programmer. Part of our next next product. He's building a big chunk of it. So we're going to announce hopefully maybe so a lot more people can play with it.

43:21

Wow. So you find with AI you want to hire younger because those kinds of people, what they use the tools and they embrace them faster. Is that the thinking?

43:33

Embrace them faster, Less assumptions. Because a lot of things you just assume you cannot do, but the assumption changes every two months, every month at this point. Right. So more optimistic, more optimistic, more hungry. A lot of experience doesn't matter anymore. You just have to be a good, ask the right questions. And I think also like you grow up with the Internet in the past 10, 15 years, then you just, you learn much faster. Like we have Internet when we were a little bit older, but this kid just completely grew up with Internet from when they were toddlers. So their rate of learning is compounding a different rate. And last couple years, if you grow up with ChatGPT, it's a compounding yet a different rate.

43:42

Support for this episode comes from Anthropic, the team behind Claude I've been using Claude to help me resolve those messy contradictions at the heart of brand strategy. How do you triangulate one message across a company's products, beliefs and future vision? Claude doesn't just tell me what I want to hear. It feeds me different strategies across a spectrum of approaches. More like a partner, less like an answering machine. When you need to get up to speed fast on something complicated, you can just point Claude at it and it pulls together sources, surfaces where people disagree, flags the things you should be asking about, all with citations. You can go check. When you connect Claude's Cowork feature to your Gmail and Google Drive, Claude will read your actual emails and docs along with you instead of waiting for you to paste things in. Then Cowork takes it further. You can point it at a folder, describe what you need done, and Claude just handles it. Queue something up before lunch and come back to finished work. Try it yourself at Cloud AI Access

44:30

the world moves fast. Your workday even faster. Pitching products, drafting reports, analyzing data. Microsoft 365 Copilot is your AI assistant for work built into Word, Excel, PowerPoint and other Microsoft 365 apps you use, helping you quickly write, analyze, create and summarize so you can cut through clutter and clear a path to your best work. Learn more@Microsoft.com M365 copilot.

45:26

I'm so happy you're advocating for toddler screen time, Ivan, because my toddler has screen time and has learned a lot from MyIPad and I'm just, I mean even dating back to the snap years, I'm just such a big believer that it's not about the screen per se, it's about what's on the screen or sure. What it's replacing if it comes to relationships, play outside, unstructured free time, all that stuff. But yeah, one would think, I mean I remember back in those days as well. I mean I'm curious, you know. I know you've. You've been with Notion a while now. I mean how would you have done it differently on the way up if you had access to these tools?

46:00

Oh man, good question. Anybody? Nobody asked me this before.

46:35

This is what you come on the show for, Ivan.

46:39

Yeah, I'm not a researcher so I don't think I will be building. Even if you rewind back five, 10 years ago, I wouldn't be building. I wouldn't be first deep diving into building AGI right. So actually you guys know there's historically there's always two spectrum of Schools of computing. Right? So on the east coast There is this AI AGI school which is like the start MIT John McCarthy trying to teach computer how to play chess. This is 1950s MIT, right? That's on the east coast. On the west coast this is like Douglas Engelbar and later Alan Kay. Those are the HCI human computer interaction where you can known as like augmenting human intellect. AHI school for computing. So AHI hi always be polar schools of computing. I would say right now the power definitely shift a lot towards AI HDI side. But what draw me to star notion draw me to start play with computers. It's the interface, the tool side, the create the shape of something that you can hold in your hand that hopefully amplify what makes us human. If I start in tech again now, I think I'll still be drawn towards that directions like in notion. Some of you been to our offices, we care a lot about tools and craft. That's the hci, that's the human augmentation schools of way of thinking about computing. And I think right now it's an interesting time because AI becomes so powerful. How do you use this power to create new forms of tools? Is the question that nobody proper answered yet.

46:41

Openclaw is a good example of that

48:22

because Alex, why do you have to incessantly bring up openclaw?

48:24

Well, it's just on the mind man. I think what he's saying is, is connected to that because I think a lot of people, a lot of labs had that kind of tech internally and AI is getting a lot better at tool use and agentic use cases. And really what it is, it's like a different interface. I mean it's not a good one in terms of like a mass consumer interface, but it just changed the interface to interact with an agent in this way and make it accessible. And so the opportunity with interfaces in AI right now is just tremendous because there's so much, there's so much ceiling on just the AI we have today even beyond future models. Right.

48:27

I think that's people start showing like hey, real time strategy games from the 90s and 2000 might be a good form factor for managing agents.

49:03

That's what I'm talking about.

49:12

We're Kanban board. Turns out we're full circle now. Like for we had a one year, two year of managing IDEs using IDE to manage agents. Now we're going back to managing multiple agents. Turns out the best way is Kanban boards. So we're still looking for the best form factors.

49:14

I was the settlers of Catan addict in college. I won the Kentucky championship one year. Are you saying why? Maybe that could come back and I could use that skill to run my agents. I don't know about catan, actually. It's probably more. It's probably more like civilization.

49:30

I got one. What is the zergling rush of agentic deployment in StarCraft terms? If you really want to destroy a competitor before they've even gotten out the

49:46

doors,

49:59

I need one agent to DDoS their servers. I need another to hire all their team members. I need another to. What?

50:01

We're missing metaphors here.

50:09

Did you play StarCraft?

50:12

I do play StarCraft. I like work. What'd be interesting to see? It's like. Like human wants to be in the loop. You just have to go up the loop. You don't want to be on the level of doing busy work loop. You want to elevate. Elevate. Eventually you're from managing work itself to managing a city. And I think we're still doing that. We're figuring out a right interface allow us to elevate.

50:14

You've inspired me to want to not look at email ever again. That's gonna be my first level of elevating.

50:40

Well, my prediction is like, there's agents to manage inbox and there's agents to write emails. And I think in maybe two years from now, email will be a really flooded medium just with all the spams back and forth. So become too saturated. Right. So we'll see. Oh, it gets really busy now with all the sales emails and all the agents.

50:46

If you think that happens to email. I'd actually, I was going to. I've been trying to find a way to ask you this. I'm curious how you think notion itself survives agents at scale? Because there's the argument that what cloud is building and what OpenAI will probably do with Peter and what they're doing on agents with Codex, like maybe they want to eat you too. How does notion stay firm in that kind of an environment? Where I was listening to Peter on Lex Friedman, the founder of OpenClaw, and he was saying most apps will just become API calls. So how do you avoid notion being in that future? Or does that even matter? Is that okay?

51:07

Yeah, I think it's a lot of discussion. Like, it's Software dead. Is SaaS dead? Right. To me, that's kind of like a tale of two cities. Like with agents with language model, like how software is used and consume and build is completely different. Like, in some sense, software providers Are you're no longer just building for humans, you're building for other language models and agents. Like if your product cannot be used by agents, I don't think the future is really promising for you. But if your product can be rearchitected, can be properly adopted, used by agents, you use the word to use. I think that you can tap into brand new market that's that the tech industry have never seen before. Like for the longest time tech mostly selling tools. In the past 15 years, selling tools as a C based business model for other people to buy. But the larger market is not just selling tools, is selling the work and services. That's about 10 times the market. So if you can make your product and your current stack and all the tooling you're building around this to tap into that market, that's much, much larger. And to me that's the opportunity for all the tech companies. The question is, can you reshape your architecture to ride this wave? That's why we have been building notion so many times. To ride this wave.

51:44

Yeah, it's crazy to think about the back end versus the front end. That's what I was wondering about. If you were to build notion from the ground up with these tools, if the user interface would have been any different. I mean it's definitely clear. You know, with Yalls incredible design sensibility and I think we talked about this once Ivan, just like that eternal unsolvable tension, you know, between too many buttons and too few buttons and how do you create some way for the team to know what's right, to know what's enough and what's too much. But it seems to me that with AI a lot of this stuff could ostensibly be hidden and a lot of the buttons could go away. I mean when you think about it, it's like moving your mouse and clicking something may be something that that 16 year old couldn't care less about versus just kind of speaking it even which you're starting to see a lot of different speaking tools that register this stuff even more.

53:05

That's why I like I've been using chat as an interface to manage my email inbox. Right. Like what's matter here is like I don't want you to click 50 button to archive emails. I can just tell the customer agents hey archive what's I don't think it's important today, right. So I think what's key here is okay, fundamentally the company kind of gets more nimble, shaped differently by future business. They were built by a small group of humans start A company together. A lot of talk about one person building a billion dollar business, but at least we believe it'll still be a small group of people. So will be some form of multiplayer, right? Multiplayer type of use cases is more important than single player. That's one form factor that's important. So when multiple people will collaborate, what's the best form factor for multiple people to collaborate? Reading and writing text is still one of the best form factor. So you need a multiplayer canvas to do some form of documents, some form of writing. And when you have a multiple people collaborate on strategy and writing, you also need to collaborate on what to do, who to do, what. That's basically give us some kind of a project management shape of things. Whether you're managing to do for humans, managing for to do for agent, you got some kind of Kanban board, right? To me those are the core primitive to work with a small group of people and delegate a lot of work to agents. Agent can write a lot of docs, agent can do a lot of tasks, you can work with external cloud code and cursor agent to do creating new feature for your code base. But that's kind of your command palette, command dashboard for human. A small group of human to work together. And lastly you want to have the state of our models. So right now there's three or four model companies and each ones leapfrog each other every couple weeks. So for business buyers they don't want to lock into any of the model providers. So there's a value to be neutral, there's value to be a switzerland of auto model providers. So you want to have a multiplayer interface collaborative environment that can use state of our models that are model agnostic and neutral and truly B2B rather than B2C. So that's the position we see that's brand new, wide open the market and that's what we're going after today.

53:59

And that's how you think notion will withstand OpenAI and anthropic trying to go hey notion, you're actually just an API call to our agent and our agent is going to traverse your interface. Right? Because that's what they would probably like to do in the long run.

56:23

I think we should open everybody because like it should be more open rather than close right now. So many company SaaS provider close down their API, close down the endpoint I think is the moment actually to be opened up. So not just welcome human to read and write your product and services, welcome agent to do that. And you can win by providing a really good Thoughtful user interfaces. You can win by being Switzerland and play with the state of our models. Like whenever openehr and Tropic release a new model the next day or the very same day, we have the model, customers want that and we can play neutral in this way. And we already have the artifacts of documents and Kanban board and to DOS and all the permissions and security in enterprise care about.

56:39

So when AI can use your apps, access your tools, use APIs, access other things through MCP, I mean it seems to me that it's kind of turning a lot of services into at least more so of a dumb pipe than they have been. And if it's kind of being extracted or sorry abstracted away. You wonder kind of what happens to these companies and the types of edges that they were able to achieve by having their own interface. Do you think that's what's behind some companies resistance to being open?

57:22

The future is really unknown. People don't know how this is going to play out. There's a lot of fog at work. We've using some of the real time strategy metaphors, right? You don't know the future, you tend to play it safe.

57:56

You don't want to accidentally drain your

58:10

moat, you don't want to accidentally drain your data, drain your sheep. Right. At least in notion we believe like when other people zig or should Zack when the entire software market sort of closed down their data, it's a time to actually embrace open, embrace developers, embrace getting tools, coding agent to access your data and just be a really good in and out, be a really good human canvas to collaborate. That has been our strategy. So beyond our next week's customization release, we're going to release a lot more developer facing features in the coming weeks and months and those are embracing this openness of the platform. Right. Notion historically hasn't been a very developer centric company, but this is about to change because this philosophy, so it sounds

58:12

like the nuanced view you have is that the whole SaaS is over market sell off thing maybe is overblown in some cases, but there are companies that if they are not designed for agents to traverse, you think they're going to have a really hard time and that's kind of going to be the pecking order is you've embraced this, you're going to be more open, you're going to let agents crawl, you be open your endpoint or you will be kind of left behind. Is that, is that fair?

58:56

That's fair. I think there is a change in terms of how fast can you rearchitecture your product? Right. So if a product has a lot of buttons, I don't want to speak poorly for anybody's product, but assuming a product that 90s, early 2000s software with like a thousand buttons on the screen, it'll be really difficult for agent to understand how to use it. And if agent cannot use it and if the majority of human knowledge work going to be done by agents, then you cannot lose market share. It's very simple. Right. Second, it's the business model transition. So you no longer just selling seats, but you should think about how do you selling usage, selling work, selling outcomes. There's a lot of buzzwords and nobody have a right answer in the market. Literally not a single company figured this out outside some developer tools. So everybody's just watching each other and. But we're going to figure it out in the next year or two.

59:24

Do you think you'll move to usage based? Because to my understanding you're mostly seat based. Right? That's the traditional.

1:00:17

That's what we're doing right now. Yes.

1:00:23

Really?

1:00:24

Yes.

1:00:25

Yeah, that seems profound and that seems like that could change the product a lot.

1:00:25

Yeah, it's a lot of experiments. So Customer Agent is our first usage based product and customer actually embracing it. Just we're talking about the ramp vercel clay cursor of the world. If you just ask the value they get from those customer agents, they're saving meaningful real human hours and that can translate into dollars in a substantially amount away rather than just seats. So people are embracing because the benefit it creates for people.

1:00:32

So I'll make you sweat a little bit. Ivan, I know you were talking about kind of using email through the AI today. Why was it important to acquire the email app and the Calendar app? Was that about having better access to some specific data, coming to own it? I know a lot of apps in the past want you to create their own pipes instead of using the Gmail pipes, et cetera.

1:01:00

Yeah, we want to create our own pipes. So we bought the Mail and calendar product two years ago and at that point we want to create our own pipes. It's still the plan long term, but there's an even faster way to create value which is to create the best tool used for mail and calendar two

1:01:22

ways in different paces. It's like sooner rather than later. We're going to have computer and browser use later. It'll just be the pipe and we'll get there both ways.

1:01:42

Yeah, you need to build good pipes, right? Like the Model capability has been there for quite some time, pretty much a year now. What's missing are the infrastructure and the pipes to do knowledge work, the high quality tool use capability that works well together, how the permission pipe together and that's what we're building. So we've been repositioning our mail and calendar team to build those pipes and make sure they work really, really well out of the box.

1:01:54

Yeah, because like why would you even need all these separate apps in the future? Why would you need a separate calendar app, email app, et cetera. That whole model seems like it's going to go away.

1:02:17

I think Calendar, email, tbd calendar, you still need a very information dense way to see things. So the interface is still really critical. You can argue if you are professional salespeople or someone lives in inbox and you see like 5, 10 emails at a time. Interface still matters for me, my at least for my personal email habits, I usually pick out three or four important email at each given time. So interface is less important. So it really depends on how you how what kind of information density you want. And calendar is really hard to beat, has been around for like hundred plus hundreds of years, this kind of way of layout. So I think calendar is here to stay.

1:02:27

Well, so how are you thinking about the product suite growing? I think I'm sure there are folks internally who are like let's just create a bottom nav bar inside main notion that includes your docs, your calendar and your email. But then there's also the school of thought that having these as separate apps for any number of reasons is pretty important. I feel like the consumer appetite is definitely for simplicity as opposed to super app, at least here in the west. So what's your take on that?

1:03:04

We're actually changing. I'm personally changing my philosophy on this. Oh, given how quick you can just build things with coding agents, hopefully soon enough we're going to create a release, a dedicated customer agent or agent chat product that has all your business first chat needs in one place. Right. And that's a dedicated use cases. That's chat first notion with all the context in notion and you can use it to manage and calendar and email. And this is a standalone app and we can build this fairly quickly because now a small team with a coding agent can just whip it out in a few weeks.

1:03:31

Ellis just started sweating with anticipation.

1:04:10

Yeah, I mean as a designer that's the number one goal, right Ivan, you want to fill up the entire iPhone doc with your own apps. Is that not the goal for every designer?

1:04:13

I want to turn them all into black and white color, mono color.

1:04:21

Oh man. I guess the only one left is slack then, huh?

1:04:26

Yes, Ivan, I wanted to ask you about the company more broadly. You all just did a pretty big tender offer, let employees sell some stock. I noticed you waved your cliff the one year cliff for new employees and I think everyone now that Figma went out, went public, people are wondering when is notion going to go public. Is notion going to go public? How are you thinking about the next phase of the company like that?

1:04:29

Yeah. To us we share this message very clearly internally. It's most important for us to layer our business model into this usage based business model like the one we talked about earlier. Not a single software company has done this at scale yet outside developer tools and people are looking where in the knowledge work domain is one of the largest if not the largest market. But not a single product has do this kind of conversion into selling work to other knowledge work customers. So we need to do a good job, do this kind of transition while we're private. So we can take a lot of risks, we can move really fast, we can watch how the public market plays out and this could happen in the next two or three or quarters. Right. And when this is figured out, when we fully release custom agent with a new pricing model, with new business model, then we can talk about IPO timing.

1:04:57

The only other person I've seen talk is clearly about that shift to usage based as you is Brett Taylor at Sierra. That's how they've been operating from the beginning and it makes sense, right? I mean he puts it very plainly like why would you not charge for outcomes if you could and you could maybe charge more. And I think people are so used to the kind of frankly lazier seat based SaaS model. It's almost like a gym membership where you probably make a lot of money off of seats that don't get used. Right. So I imagine especially in AI which is expensive to serve and you guys are a huge token buyer from the labs already. I imagine there's a lot of trade offs to going to that approach. But it seems like you have conviction that this is going to be the future and that every business like yours will eventually move to this and you all want to be be early to that, right?

1:05:51

Yes, because the market's much, much larger. Right. Selling tools, selling seats, you're still selling, you're still in the tech industry when you can selling work and usage and outcome, you are in the knowledge war economy. That's ten times larger. Right. And that's, I will say that's the direction that more companies, more software companies should look towards to not think about, okay, can someone wipe code their own products and they just shrink the cost of creating software. That's more interfacing margin looking but outer looking one is like there's entire knowledge world economy. There's so much busy work in the world. How do you help people delegate some busy work? That's a much larger market, a more interesting market.

1:06:41

I know there have been a lot of different ways companies have been positioning themselves and you just bring up busy work again. I mean how are y' all working to differentiate yourselves at this point? I know there's kind of been the teammates angle from notion, but it seems like when a lot of folks are talking about the same thing, it's got to be pretty tough. I mean especially to enterprise customers. What are you guys saying these days? What's the latest and greatest?

1:07:27

Yeah, if you think about the market, who are the players in this kind of knowledge workspace? We mentioned Open cloth, right? Opencloth is Mac Mini, very much for tinkers. You can manage your personal knowledge or personal work with it. That's one space. It's a single player. It's fairly complex to use but quite powerful sitting on one of the quadrant. And here we also have Claw Codework which is claws sister product for cloud code, tailored for knowledge or less for coding. And it works great with files, right? Lawyers love it, spreadsheet finance team loves it. And entropy really doubled down on files. So there works really good for all the formats, known formats, but because of the files it's still largely single player. It's good for individuals. I think the niche that we can occupy is this multiplayer from the get go sits on the cloud and it's made for business, made for teamwork, not for personal use from the get go. And that's a niche that not that many players are going after today. And we have a head start because we've been building a collaborative product for more than five plus years. We have all the canvases and building blocks from a document for project management tools. We've been building AI product for more than two, three years. So a lot of pieces are there and allow us to ship the customer agent as one of the first in industry to do those kind of things for business. And that's our position, what we want to do really. We really want to nail for the next coming quarters.

1:07:53

A lot of talk about multiplayer, but I still can't see people typing live in notion. That's the only thing I think Google Docs still has is that like a architecture thing.

1:09:30

We have the architecture to build it. The question is just like do we spend more time on the AI stuff or do we go back to do the Google Doc competition side of things?

1:09:44

Just hire a couple more teenagers. You'll get it done in like a day.

1:09:54

If you have your cousins or your nephews to recommend, we're happy to take them.

1:09:57

While I have you. I feel like, you know, when you're on a meeting and presenting or working at the same time with people, it's maybe not as much watching them type, but being able to highlight specific things. Like look at this. You know what I'm saying?

1:10:03

You see cursor and C highlight is the same architectural or implementation complexity. So I'll add one more to do for.

1:10:16

Or do you have an agent that is going to review the meeting notes? Are you running Notion AI right now? I am not meeting us.

1:10:25

Of course you are.

1:10:32

I'm not actually. So.

1:10:33

Okay.

1:10:35

Wow.

1:10:35

What have been smart to do.

1:10:36

I assume you guys will have recordings so I can just.

1:10:37

I mean.

1:10:40

Yeah, but you need it for your agent's context. How's it going to know what to tell you tomorrow morning?

1:10:41

I think we're mostly covering the things we already talk about. You guys to ask better questions. Right. Auto.

1:10:47

We didn't cover anything new. Don't need my note taker. That's a great modern burn.

1:10:55

This isn't a new one. But I did want to know what it felt like to get quoted alongside Steve Jobs and Bill Gates by Satya Nadella at Davos. I know you saw this. He really loved your Manager of Infinite Minds. You wrote this essay that was really great. It got shared a lot around the holidays about how you're thinking about AI and this, this phrase, manager of Infinite minds, which is the kind of phrase Ellis would spend days in the shower trying to come up with for a founder. And Satya was giving you mad props for that. I have to imagine that was pretty wild to see that.

1:11:00

Yeah. Some bunch of people text me in live. I'm very flattered. Yeah. Microsoft is a company we look up to. Like it's one of the company that put PC personal computer on every desk, every home. Right. So it made this industry and Steve Jobs, those people I look up to when I got into tech. And I think Satya is the real one because he actually came up based on what I know through the rank of Excel building all the. Basically no co product of the 90s. Right. OpenDoc Olay type of thing. So he knows what he's talking about and see him rephrasing something we published during the break. So it's a huge honor.

1:11:34

Manager of Infinite minds though. How did you come up with that? Ellis needs to know.

1:12:22

I'm actually not sure. It could just be notion agent regurgitated a thousand times and some words just show up. Then I probably play more of editor roles. Oh wow, that feels right. So I kept it. I like to think in metaphors. Metaphor just like help me think through problems from different angles and also other people can relate it to. So like Bicycle for the mind is Steve Jobs quote, right. But if you think about that and we have Internet in it since the 90s, we're still pedaling bicycle on information highways. So when I express this metaphor to my team and for other people, people get it. So what's coming? It's like we can drive cars on information highway through AIs and people get that. Those are pretty powerful things.

1:12:26

Speaking of word choice, I know something we've discussed in the past Ivan is workspace. When that 16 year old or that 18 year old shows up. I mean obviously they're not your icp. But this is something that Microsoft and Google have been using for years. Is that still the right word? How are people going to talk about this in the future?

1:13:10

Everybody start using the word workspace. I don't love that word that just become what it is. The meaning behind it is still important. I think it's still going to be around which is you have a centralized place, a system record of things, a system record of workflows. The power your project or power your business and it will be a group of people looking at the same thing together. I think that will be around if anything become more important in this kind of agent first world, right? And how do you allow that get different agent to access to this workspace? You invite cloud co agent, cursor agent coming to based on your understanding what your business trying to do and start working on your workspace in GitHub and do something. There's this notion still be around. I think what's different is in the past is the human doing a lot of information pushing from this workspace to the other workspace which is a majority of knowledge work. Now we have this magical materials of infinite minds who can do this for us. And how do we elevate ourself to not just managing one workspace? Can we manage intent? Can we play a sim City of 1,000 workspaces at the same time? Maybe it's possible. And the limiting factor again is human interface, in my opinion.

1:13:31

Here's bringing it full circle. My genius idea. If we want to appeal to all the strategy bros, we could just call it the war table where you direct all your. We direct all your field troops to destroy your enemies. Or on the other hand, serve your customers with delightful solutions.

1:14:49

Mm. Put into cloud or notion AI. Find better options.

1:15:08

No, Ivan, just think about it. Just think about it.

1:15:14

Okay, I'll keep thinking about it.

1:15:16

We should let Ivan go. Ivan, we appreciate your time.

1:15:18

It's a pleasure, gentlemen.

1:15:20

Thank you for joining us. For joining us.

1:15:22

I'll see you next time.

1:15:23

Yeah, see you next time. Bye.

1:15:24

And that's it for this week's show. Don't forget to like subscribe everywhere you get podcasts. Our website is Access show and you can find us on video at access pod on YouTube.

1:15:29

Thanks to Ivan from Notion for coming on this week. And if you liked this episode, please leave a five star review. All five stars or a thumbs up. Share it with a friend. Really makes a difference. And you can find my newsletter at Sources news.

1:15:40

Hashtag Sources News. And you could find me at Hamburger on twitter and@meaning.com Access is part of

1:15:55

the Vox Media podcast network and the show is produced by by Hooked Creators.

1:16:02

Bye. Thanks again to Anthropic, the team behind Claude for supporting this episode. Anthropic is committed to keeping Claude ad free. Your conversations in there won't be shaped by advertisers, which given all the details of your life you share with AI is worth paying attention to. And for the people who want to dig deep into something, whether that's writing research or just working through a hard process problem Claude is worth a shot. See why problem solvers choose Claude as their thinking partner? Try it yourself at Claude AI Access.

1:16:06