Notion Ceo Ivan Zhao Wants You To Demand Better From Your Tools

Hello, and invited to Decoder! This is Casey Newton, laminitis and editor of Platformer and cohost of nan Hard Fork podcast. This is nan 2nd section of my productivity-focused Decoder bid that I’m doing while Nilay is retired connected parental leave.

Today, I’m talking pinch Notion cofounder and CEO Ivan Zhao. I’ve followed Notion for rather immoderate clip now — I’m a large fan, and a awesome portion of my workflow for Platformer is really built connected apical of Notion’s database feature. So I was very excited to get Ivan connected nan show to talk his accuracy connected productivity, really he’s grown his institution complete nan past decade, and wherever he sees nan abstraction going successful nan future. 

If you’ve ne'er utilized Notion, you tin deliberation of it arsenic an all-in-one productivity suite comparable to a batch of nan collaboration and alleged “second brain” apps connected nan marketplace — from nan much business-y task guidance devices for illustration Asana and AirTable to nan much DIY note-taking variants for illustration Anytype and Obsidian. 

Notion sits beautiful comfortably successful nan mediate here, since it tin do what a batch of these kinds of apps do very well, and each successful 1 package. At nan aforesaid time, it allows for a beautiful important magnitude of customization, which has made it celebrated some for individual productivity powerfulness users and for companies ample and small. 

But Notion started arsenic a very different portion of software, and its improvement complete nan past 12 years aliases truthful has progressive a adjacent magnitude of proceedings and error, 1 awesome reboot, and a batch of large decisions. In my opinion, what really sets Notion isolated from truthful galore of its peers is Ivan’s heavy passion for creation and an almost relentless thrust to make products that he sees arsenic useful and artistic successful adjacent measure. 

You’ll perceive Ivan early connected successful this speech reference LEGO: nan artifact bricks are nan cardinal inspiration for Notion, which employs “blocks” arsenic a metaphor for configurable templates that let you to usage Notion successful a beautiful divers group of ways. Think everything from elemental notes and lists to analyzable databases and workflows. 

But Notion, for illustration truthful overmuch package these days, is evolving. Now, nan institution calls itself nan “AI workspace that useful for you,” and you’ll perceive Ivan recount successful item really nan motorboat of OpenAI’s GPT-4 proved to beryllium a large turning constituent for him and for Notion. The institution launched an OpenAI-powered AI merchandise overmuch sooner than nan competition, moreover earlier nan motorboat of ChatGPT, and it’s added a big of caller AI-powered features successful nan past fewer years. 

Ivan himself is besides beautiful excited astir nan capabilities of AI; he said he uses it successful his free clip to study astir caller subjects, and you’ll perceive talk successful extent present astir his imagination for AI agents that progressively do much and much activity for you wrong of apps for illustration Notion. 

But a communal taxable pinch nan AI manufacture correct now is nan very ample spread betwixt what AI tin really do today, and what truthful galore group dream it tin do down nan road. So I really wanted to inquire Ivan really we mightiness get to this early he predicts, really agelong that will really take, and what productivity and knowledge activity look for illustration if AI delivers connected immoderate of these lofty promises. 

Okay: Notion CEO Ivan Zhao. Here we go.

This question and reply has been lightly edited for magnitude and clarity. 

Ivan Zhao, you are nan co-founder and CEO of Notion. Welcome to Decoder.

Thanks for having me.

So astatine a precocious level, picture Notion for us. If listeners haven’t utilized it yet, what is it? What does it do?

Well, we make all-in-one productivity software. People usage Notion for each kinds of things, from taking notes, collaborating connected projects, managing documents, managing their knowledge base, and, astir recently, we launched a almanac merchandise and message product. You usage Notion, truthful you should picture what Notion is.

I deliberation you conscionable did a really bully occupation describing it. I do usage Notion, which is 1 of nan reasons I wanted to talk to you, because each clip I talk to a CEO whose products I use, I get to springiness them merchandise feedback, which is breathtaking for me.

Perfect.

So do you spot Notion coming arsenic much for teams than individuals? Is that nan guidance it has found?

We designed Notion for teams. So different measurement to picture Notion is that we telephone it LEGO for software. Maybe it’s worthy explaining nan intent a small bit. If you’re a company, for your team, you person to usage a twelve different devices to get your activity done, and our extremity is to consolidate those devices into 1 container and springiness you nan LEGO blocks that powerfulness each those usage cases. Not only tin you do each your activity successful 1 place, but you tin besides usage those LEGO blocks to create and customize your ain workflows.

You and I person talked a adjacent magnitude astir LEGO complete nan years. What appeals to you astir that design? Why is it specified a bully metaphor for what you’re trying to do pinch Notion?

Well, because it didn’t rather beryllium pinch software. If you deliberation astir nan past 15 years of [software-as-a-service], it’s mostly group building vertical constituent solutions. For each buyer, for each point, that solution benignant of makes sense. The measurement we picture it is that it’s for illustration a difficult integrative solution for your problem, but erstwhile you person 20 different difficult integrative solutions, they benignant of don’t fresh good together. You cannot tinker pinch them. As an extremity user, you person to jump betwixt half a twelve of them each day. 

That’s not rather right, and we’re besides inspired by nan early computing pioneers who successful nan ’60s and ’70s thought that computing should beryllium much LEGO-like alternatively than for illustration difficult plastic. That’s what sewage maine started moving connected Notion a agelong clip ago, erstwhile I was reference a machine subject insubstantial backmost successful college.

You wanted to make devices that would threat together nan measurement that LEGO blocks do.

We want to make devices that amplify quality creativity. LEGO are creative. LEGO are beautiful, and astir package astir apt is not arsenic much.

Say a small spot much astir really you were drawn into this world. Were you ever personification who was willing successful productivity tools? Or did that travel to you later successful life?

No. I deliberation it’s a misunderstanding astir Notion. Notion is productivity software, that’s what we do arsenic a business, arsenic a product. But nan ethos of this is what I conscionable described; it’s nan LEGO ethos. It’s possibly worthy describing nan history of nan computing manufacture a small bit, because that inspired Notion. 

Yeah. 

In nan ’60s and ’70s, ’60s, nan hippie procreation who took acerb successful San Francisco thought, “Well beatified shit, this room-sized calculator, if you put a show successful beforehand of it, it tin beryllium an interactive thing, it tin beryllium a caller type of mean that helps you deliberation better, helps you lick problems much collaboratively.” That’s why nan first procreation of individual computing started successful nan Bay Area.

That procreation of thinkers and pioneers thought astir computing benignant of for illustration reference and writing. Like we went to schoolhouse for aggregate years, but not everybody tin publication and constitute English aliases German aliases immoderate connection you speak. Writing is simply a tool. Yes, you tin beryllium a poet, you tin beryllium penning essays, but it’s a very malleable medium.

So, they group retired to make computing malleable and tinkerable, and everybody could their ain software. Then, successful nan ’80s, nan Bill Gates and Steve Jobs procreation took computers to nan wide marketplace and put a machine successful each home, connected each desk. They benignant of froze computing into this exertion format. If you deliberation astir applications, each exertion is benignant of for illustration a mini-prison of computing. You cannot alteration it that much. There are exertion makers who are technologist programmers, and past location are exertion users who are for illustration nan remainder of us, nan group who usage productivity devices each day.

So erstwhile I was reference those papers astir those ’60s and ’70s people, I thought, “Holy shit, nan world that we’re surviving successful is for illustration a prison-like world.” If anything, nan SaaS of nan past 15 years has been for moreover smaller situation cells, arsenic each exertion tin only do a mini portion of things. So, that doesn’t make it correct for me, and nan customer feels nan aforesaid way. It doesn’t make consciousness that for your regular job, you person to jump betwixt 20 different devices to get immoderate activity done. The mean institution aliases mean business uses 100-plus different SaaS tools. The fragmentation is obvious, moreover for nan IT department. 

So, there’s different saying successful business: you either bundle aliases unbundle. So, Notion’s squarely successful nan bundling business. Our occupation is to bundle SaaS into a one-ish productivity instrumentality for your halfway regular needs, truthful we tin unleash LEGO-like productivity for you.

It’s an absorbing conversation. It makes maine deliberation of room gadgets, because you spot nan aforesaid hostility there, wherever location are immoderate room tools, for illustration I don’t know, a guidelines mixer, aliases an immersion blender, that you tin usage to make many, galore different kinds of recipes. And past there’s nan ail press, which is bully for mincing ail and thing else. 

It sounds for illustration what you’re saying is by nan clip we sewage into nan 2010s, productivity was conscionable a bunch of ail presses, and you benignant of wanted to travel on and say, “What if we conscionable had a guidelines mixer and you could make a batch of recipes pinch 1 thing?”

A friend of excavation utilized this metaphor, akin to what you said. Have you seen avocado cutters?

Yes.

An avocado cutter is made conscionable for freaking avocados. You cannot do thing else. In comparison, a room weapon is simply a instrumentality that you tin usage hundreds aliases thousands of different ways. You, arsenic a human, amplify it because you person a technique. So, to create package that’s much for illustration a room knife, aliases LEGO, that’s what interests me, and interests america arsenic a company.

But you can’t blasted nan industry, because if you deliberation astir it — if you rewind backmost earlier SaaS — nan world was moving connected Microsoft Office for a bully coagulated 10 to 20 years. SaaS, pinch nan net arsenic its distribution, allows caller businesses to beryllium created. The earthy measurement to spell astir distribution and caller businesses is to find a really precise solution, creating those avocado cutters and ail presses, and you tin find buyers connected nan different side.

So, arsenic we move into today, do you deliberation of yourself arsenic competing straight against Microsoft Office aliases Google Workplace? Is nan imagination that big, aliases is it thing different?

We coexist pinch them, arsenic astir of our customers are still utilizing Google Workplace aliases Microsoft Office. They usage their personality service, they usage their message and almanac apps. We person a message and almanac merchandise presently arsenic a client. A startup tin afloat tally connected Notion. You don’t request to usage Microsoft 360 aliases Google Docs, but it’s not arsenic mutually exclusive.

Our saccharine spot is much connected nan things that you request to put successful a database. Another measurement to deliberation astir it is like, what is Microsoft Access but for nan 2020s, and AI native? Most SaaS is benignant of for illustration a relational database, storing immoderate benignant of strategy grounds of your company, and 1 workflow connected apical of that. That’s nan portion that neither Microsoft nor Google touches today. 

There are spreadsheets, but location are not galore database usage cases. We want to consolidate and commoditize that, and springiness group nan LEGO of those database usage cases, specified arsenic task guidance and summons tracking. Some companies usage them for CRM, aliases managing exertion trackers. For reporters, you tin negociate each your leads and nan stories. Those are database usage cases.

Right, and I do do immoderate of that successful Notion. Let maine inquire you astir nan flip broadside of building a merchandise that has truthful overmuch inferior baked into it, which is that sometimes erstwhile I’ve talked to group who person tried Notion, they say, “I didn’t cognize wherever to begin. I felt intimidated by nan blank page.” It seemed for illustration location was a learning curve.

How do you deliberation astir that challenge, and effort to bring group on into knowing what Notion is meant to do?

Yeah, for illustration early LEGO, you get bricks. Then later on, LEGO created systems and boxes, and now LEGO useful pinch Marvel and F1 to create really specialized boxes. In immoderate sense, Notion arsenic a company, we’re successful nan mediate of adding much boxes, truthful group don’t person to commencement from an quiet group of bricks, pinch nary instruction manual. 

You tin imagine, like, “Hey, I want a Formula One title car. I for illustration that LEGO box.” When you unfastened it, you person your car ready-made for you, truthful you tin commencement driving it. You play pinch that LEGO artifact correct away. But if you don’t for illustration definite parts of nan cars, because they’re made from LEGOs, you tin alteration them. That has ever been our philosophy, and we’re doubling down connected this attack because it works.

Interesting. I consciousness for illustration a cardinal situation that immoderate of nan different large productivity devices for illustration Microsoft Office person had complete nan years is bloat, right? The app has a cardinal features, and each individual 1 is very important to for illustration 0.5 percent of nan personification base, truthful you can’t region it.

But also, nan app conscionable gets harder and harder to usage complete clip because it’s truthful stuffed pinch buttons, menus, and widgets. Can Notion debar that? And person location been times erstwhile you’ve worried you mightiness beryllium location already?

It’s decidedly tricky. If you want to support much power, you request to person much things. There are 2 ways to attack it. The classical measurement is conscionable adding that feature, successful nan difficult integrative way. We’re taking a much LEGO approach, truthful adding nan brick, and nan ceramic tin beryllium utilized for different things. In immoderate sense, this is simply a overmuch amended approach.

Got it. So, trying to connection less discreet, very constrictive features, and much absurd features that tin beryllium extended successful various different ways.

Like LEGO systems, connected 1 extremity tin beryllium artifact cars, and connected nan different could beryllium Barbie dolls, much aliases little utilizing nan aforesaid bricks. If you look astatine nan astir communal productivity tool, if you conscionable put your designer mind connected that, location are 20 aliases 30 pieces there. There’s immoderate benignant of table, immoderate benignant of relational database feature, immoderate charts, immoderate commenting, page editing, and collaboration. 

Those 20 things are halfway to each collaboration and knowledge work. So, we effort to do our champion occupation to make them friendly, approachable, and break them into pieces, and springiness them to you, either arsenic a portion aliases arsenic a portion of a package.

What bricks wrong of Notion are astir celebrated today?

We commencement pinch bricks astir documents and knowledge bases. We’re celebrated for our block-based editors, and that’s from nan early days. That’s for illustration 2019 into 2020. And past databases are our astir important ceramic today, because, for illustration I mentioned, astir knowledge activity is conscionable fancy record cabinets successful nan cloud. Knowledge activity runs connected record cabinets, and databases are nan bosom of that.

Yeah, nan database is my number 1 Notion ceramic that I use. So that makes consciousness to me.

People don’t observe that [easily]. We request to do a amended occupation of getting group to understand its power. It’s fundamentally what an technologist does each time is ligament together a relational database pinch views connected apical of that. How do we democratize that? That’s our purpose.

Well, it’s interesting, because if I had ne'er heard of Notion, and you came to maine and you said, “Casey, you should build a database to lick this problem,” that’s for illustration telling maine that I should adhd different room to my house. I don’t cognize wherever to begin. I consciousness for illustration I request to telephone personification and inquire for help. 

But successful practice, you click a mates of buttons, and successful my case, instal nan Notion Web Clipper, and I’m good connected my measurement to having a database. So, I don’t really deliberation nan learning curve is that steep, but I could understand why personification mightiness beryllium intimidated by it if they had ne'er tried to do that benignant of thing.

Yeah, not each kid grew up liking LEGO arsenic their number 1 toy. I deliberation Notion resonates nan champion pinch group who for illustration to build, and they thin to beryllium entrepreneurs, tech people, and nan spreadsheet gurus successful each team. They for illustration Notion, and they group it up for nan remainder of their teammates.

That’s ever adjuvant erstwhile you tin get group wrong nan institution doing nan income portion for you.

You cognize what? AI tin do this rather good now, because what AI is bully astatine is gluing together LEGO bricks. AI tin constitute code. Writing codification is conscionable different measurement of gluing together your process and workflows, and our latest merchandise fundamentally gets AI to beryllium this successful personification to thief group up your Notion workspace for you, and that’s different measurement to onboard customers. That’s a marque caller measurement to unlock that we’ve added successful nan past twelvemonth to 2 years.

I will opportunity that it has been very powerful for maine successful a batch of different products. Being capable to usage nan in-app AI to say, “How do I do this?” And really getting an answer. As personification who has spent a batch of clip successful thief menus complete nan years, digging astir and not uncovering what I was looking for, that has been ace useful.

And it’s not conscionable helping by school nan quality to do it. More and more, AI tin conscionable do it, right?

Right.

That’s nan biggest difference, actually. If you deliberation astir what’s happening successful package correct now, package is mostly group providing nan devices for humans to use, and much and much companies are realizing, “Wait a second, we person this caller point called a connection model. It’s for illustration a quality mini-intern successful a box, and we should creation our package to thatch AI really to usage it truthful humans tin inquire AI to do nan activity and usage nan tools, and humans tin do measurement much things pinch it.”

I want to inquire you immoderate of nan Decoder questions that Nilay would inquire if he were here. Notion was past weighted astatine $10 cardinal astir 4 years ago, erstwhile you raised your past information of funding. What has allowed you to support increasing without raising much funding? Are you profitable?

We’re profitable. So, profitable, increasing fast; nan business is doing well.

Nice. How does that feel?

It feels good. I would opportunity nan larger driver of our mundane activity is nan truth that nan package manufacture is wholly changing pinch AI. It feels for illustration nan AI era of nan past 2 years conscionable makes nan SaaS era consciousness for illustration sleeping days. It’s a bigger driver of our execution strategy than conscionable moving a profitable business.

I’d for illustration to perceive much astir really that is working. Is it nan lawsuit that executives spot AI changing various workplaces, and they think, “We request to fig retired our type of this,” and truthful they travel to Notion to thief them fig it out? 

Or is it that your merchandise teams are truthful excited astir nan possibilities that you’re conscionable now seeing them build features, which are past drafting successful caller customers?

I would opportunity customers are lagging down astatine nan moment. Most group don’t know. It happens pinch each caller technology. You don’t cognize what to do pinch it. The customer is not going to show you. It’s nan group who play pinch this, build things, and possibly person an imagination a fewer years into nan future, aliases moreover a fewer months successful nan future, astatine this point. AI is changing truthful fast. So a batch is from ourselves, conscionable playing pinch AI and realizing, “Holy moly, this is simply a very different thing. You tin lick problems that you couldn’t lick earlier pinch classical software.” Now, what are you going to do pinch it?

There’s really an absorbing story. My co-founder, Simon Last, and I received early entree to GPT-4. So this is for illustration precocious 2022, a small spot earlier everybody else. We thought everybody other would get early entree to this. We thought, “Holy shit,” because compared to GPT-3, GPT-4 is simply a brand-new thing. It’s for illustration it has existent intelligent reasoning successful it. So we locked ourselves successful a edifice room for astir a week, and conscionable tried to unreserved retired nan first Notion AI product. We really launched a period earlier ChatGPT happened. We were excited astir what you tin do pinch this caller type of material. That’s for Notion, that power comes from there.

How galore labor do you person complete there? How large is Notion today?

High 3 digits. So, 900, possibly approaching 1,000.

How do you deliberation astir institution size? Do you spot a world wherever location are 5 times that galore employees? Or do you want to support it location astir wherever it is correct now?

I deliberation there’s nary correct reply for nan absolute number, but there’s an reply for nan density of nan talent. The denser nan better.

You for illustration having fewer, but much talented group arsenic opposed to much people?

I for illustration to person fewer, if we tin do nan activity pinch less people, and there’s little connection overhead. People person much ownership, and group tin activity things crossed boundaries. That’s conscionable amended overall. The institution moves faster. The mini car tin move corners overmuch amended than a large car, and we ever telephone Notion a mini bus. We effort to support nan autobus arsenic tight arsenic possible.

What’s your org chart? How is Notion organized?

Fairly classic. It’s maine and my co-founders, Simon Last and Akshay Kothari. Simon is still coding each day. Akshay runs our merchandise and creation org and research. Our main exertion officer, Fuzzy Khosrowshahi, runs each of nan engineering. Our main gross officer, Erica Anderson, is responsible for sales, marketing, and user experience. And we person our main finance officer, Rama Katkar, and wide counsel, Hasani Caraway. That’s our classical org chart.

So you didn’t consciousness nan request to reinvent nan instrumentality location aliases do immoderate innovating, conscionable benignant of create classical institution divisions and fto group spell do their thing.

Classic institution divisions and high-quality group support nan autobus tight. So that allows you to beryllium profitable. 

How do you make large decisions? Do you person a model you use? Or is each determination different?

Well, location are nan emblematic one-way doors and two-way doors. With a one-way door, you effort to move fast, and pinch a two-way door, you deliberation a small spot much cautiously and slumber connected it. Those are reasoning accelerated and slow types of things. I’m beautiful detailed. I for illustration to activity connected nan notes, truthful there’s a definite problem I’m bully at. Personally, it gives maine energy, and I’m willing successful moving connected nan crushed successful nan trenches pinch everybody. 

There are definite parts, though, for illustration I cannot tally our finance team. Our CFO, Rama Katkar, is really amazing. She takes attraction of that. But for definite things — for illustration creation and product, engineering, marketing, and branding — I for illustration to get involved.

You’ve ever struck maine arsenic a merchandise CEO. I deliberation from nan first clip I met you, it seemed to maine what was astir absorbing to you astir your institution was nan instrumentality itself that you were building, arsenic opposed to nan marketplace opportunity, aliases thing for illustration that.

I built Notion because I wanted to build Notion, not because I wanted to commencement a institution aliases business. I wanted this point to exist.

Let maine inquire you 1 much determination question, astir 1 of nan bigger decisions you had to make. So successful 2015, you decided to unopen down nan 1.0 type of Notion, relocate to Japan, and yet relaunch Notion 2.0, which is benignant of nan Notion that we deliberation of erstwhile we usage it today. How did you make that decision?

Well, you person to, different you die, right? At that point, it was like, “We’re building connected nan incorrect thing, nan incorrect foundation,” and you cognize what nan correct point is, but it’s conscionable going to return you possibly a twelvemonth to a twelvemonth and a half to build nan correct thing. We were a institution of astir 5 people, and we were going to tally retired of money. 

The only measurement to do it was to shrink it backmost to conscionable maine and my co-founder, Simon. So, we started over. Japan is simply a bully spot because it is inexpensive, and we person ne'er been there. It was interesting, and we could conscionable attraction connected building.

I cognize different founders who person been successful that situation, and that’s nan infinitesimal wherever they gave up, because they thought, “You cognize what? Maybe I could deliberation of different point to build here, but it seems exhausting. It’s going to return a year. I’ve already put a twelvemonth and a half aliases truthful into this app. I gave it everything I had. It didn’t work.”

What was it that made you say, “No, we’re going to support going connected this. We deliberation that location is simply a imagination present that we tin really achieve”?

The extremity was ne'er to commencement a business, for illustration I mentioned. The extremity was to build this thing, and nan point didn’t rather exist. Notion is 1 of nan fewer bundling, consolidating package productivity devices retired there, and it didn’t rather beryllium astatine nan time. Software for LEGO doesn’t exist. So if I started a company, I wanted to do nan aforesaid thing. Why don’t I conscionable reset and spell backmost to Simon and maine truthful we tin agelong nan money a spot longer?

Actually, I conscionable sewage backmost from Kyoto past week, wherever location was a tech event, and nan Kyoto politician and I did a fireside chat and talked astir this story. They wanted to talk astir utilizing Notion arsenic an illustration of really we tin blend tech and Kyoto’s trade tradition. Because we’re besides inspired a batch by nan trade group successful Kyoto, really they dedicate their clip to building something, and not conscionable for money aliases fame.

That is truthful crazy. I was really successful Kyoto past week, too. I was connected picnic and went location for nan first time, and I had an unthinkable time. Kyoto’s amazing. 

You get nan consciousness that it’s a small spot slower pace. People attraction astir what [they’re making], right? People genuinely care. That’s nan thing. It’s nan main thing. It’s not nan business, it’s not their different surroundings.

Oh absolutely. I mean, you spell to these temples that are 1,000-plus years old, and nan attraction and nan craftsmanship that they put into them is genuinely inspiring. It’s profoundly beautiful. It’s very connected to their spirituality, and their religion, culture, and history. So I could understand why a laminitis would spell location and return a batch of inspiration.

Yeah, because it’s a spot slower too, you tin attraction connected virtual spaces, connected computers.

Yeah, it’s not for illustration San Francisco, pinch our go-go nightlife, our Waymos, nan statement scene, each of that.

Or New York, even, moreover much of that.

One much of these. This isn’t strictly a Decoder question, but it spiritually feels for illustration a Decoder mobility that I wanted to inquire you. What is nan champion productivity instrumentality that you usage that is not made by Notion?

I for illustration those chatbot products: ChatGPT, Anthropic’s Claude. It’s rather amazing, particularly erstwhile talking astir features, for illustration nan speech mode. I emotion those. I for illustration to study from them.

You for illustration sound mode?

The sound mode, yeah. It helps maine study a batch of different things. When I’m making java and waiting for nan h2o to boil, I tin conscionable talk pinch this point for a small bit, for illustration for 2 minutes. It’s perfect.

What do you inquire it about? What do you for illustration to study about?

Oh, each kinds. What’s astir recent? In Japan, I was reference a book astir Marshall McLuhan. 

The media theorist?

Yeah, media theorist and theologian. Many of his concepts are difficult to interpret, truthful it’s amended to conscionable activity done them pinch a connection exemplary that will thief guideline you. It’s nan champion tutor, truly. Education should beryllium very different. Hopefully, it will beryllium very different a fewer years from now.

I deliberation so. This isn’t rather a tutoring usage case, but I conscionable person to say, erstwhile I was successful Kyoto past week, we were successful nan vicinity and had immoderate clip to kill, truthful I conscionable opened Google Maps, and it benignant of opens to wherever we are, and I took a screenshot. I conscionable sent it to ChatGPT, and I said, “Tell america a spot astir this neighborhood.”

It gave maine a history of nan neighborhood. It told maine astir nan cool restaurants, cool java shops, a museum, and places that we could locomotion to. I mean, it genuinely was arsenic bully arsenic I tin ideate getting from immoderate guide, and it was arsenic elemental arsenic uploading a screenshot. The full point took 15 seconds. It was chaotic to me.

Yeah. One usage lawsuit I have, if I spell to a celebrated architectural building, is that I say, “Hey, I’m astatine this place, show maine astir it. I’m looking astatine this part, a area of this building. Tell maine much astir why that is nan case,” right? Because if it’s celebrated enough, it’s astir apt portion of nan corpus of training. So, nan connection exemplary knows astir it.

Yeah.

You tin conscionable return a genuinely guided tour, and you don’t request different person; you’re conscionable talking pinch your machine.

Well, that seems for illustration a bully segue into Notion and AI. We’ve talked astir what Notion is, really it’s changed. Notion now bills itself arsenic nan “AI workspace that useful for you.” So, what does an AI workspace mean to you? What do you want it to beryllium for us?

If you deliberation astir our strategy during nan SaaS era, it’s bundling and consolidating a bunch of different devices for knowledge activity into 1 place. What’s changed successful nan past mates of years is that it now has each that soft knowledge of LEGO successful Notion; you tin not only supply nan tools, but you tin besides combine them arsenic your AI teammates. They tin do nan activity for you. 

We are fortunate to person that knowledge activity LEGO successful 1 spot truthful you tin portion it together successful a very absorbing way. Because 1 extremity tin return notes for you, nan different extremity tin thief you manage, triage projects, and constitute documents. Those are basal things, but much and more, pinch much LEGO and smarter models, you fundamentally prosecute Notion arsenic your AI teammate. That’s nan early that we’ve been building, aliases building much toward.

I retrieve 1 clip I was gathering pinch you, and you conscionable launched immoderate of these AI tools, and you were showing maine that Notion AI was taking notes astir various meetings. So you were capable to dip into meetings astatine nan institution that you did not personally attend, and conscionable benignant of quickly drawback up connected what your coworkers were talking about.

I thought, “That’s ace interesting.” That’s nan benignant of characteristic that I tin ideate a batch of CEOs wanting, but earlier this point, they haven’t had that level of visibility into their ain company.

We conscionable launched 3 abstracted products a mates of months ago, for illustration Notion AI for Work, including nan adjacent type of this endeavor hunt merchandise you’re talking about. So, on pinch that was nan motorboat of our AI Meeting Notes product, truthful each your meetings tin beryllium recorded and transcribed. 

So essentially, your institution has a corporate encephalon of what’s going on, and you person each nan AI knowledge workers connected apical of Notion to thief you transcribe nan gathering notes and reply immoderate questions you whitethorn have. It’s rather absorbing what you tin do pinch nan exertion now.

Of each nan AI features that you’ve added truthful far, which ones do you personally find nan astir useful?

I usage AI gathering notes. Almost each meeting, isolated from this one, I record, and I usage that for meetings truthful I tin stock notes pinch different people. I usage it for myself, arsenic a starting position to dump my thoughts. And truthful I tin later retrieve to inquire AI to move nan transcription into writing. English is my 2nd language, truthful I’m not nan fastest writer, but AI tin do amended penning than I tin if I conscionable dump retired what’s connected nan apical of my mind done nan AI transcription feature.

That’s really interesting. There’s a batch of talk correct now astir AI, and whether it mightiness switch workers, aliases full workflows, aliases functions wrong an organization. You’ve talked coming astir AI being capable to service arsenic a benignant of teammate. 

Do you deliberation that AI and Notion will get to a constituent wherever executives will prosecute less people, because Notion will do it for them? Or are you much focused connected conscionable helping group do their existing jobs?

We’re really putting retired a run astir this, successful nan coming weeks aliases months. We want to push retired a much amplifying, affirmative connection astir what Notion tin do for you. So, ideate nan advertisement we’re putting out. It’s you successful nan center. Then, pinch a instrumentality for illustration Notion aliases different AI tools, you tin person AI teammates. Imagine that you and I commencement a company. We’re 2 co-founders, we motion up for Notion, and each of a sudden, we’re supplemented by different AI teammates, immoderate taking notes for us, immoderate triaging, immoderate doing investigation while we’re sleeping. 

So, each of a sudden, we’re a squad aliases a institution of 10 people. Then nan startup tin tally overmuch faster. That’s nan imagination we want to push much towards nan world. So, much of an amplifying force, alternatively than a zero-sum force.

What timeframe do you deliberation that arrives? Does that consciousness for illustration it’s almost wrong reach? Or do you deliberation we’re going to request to spot respective much investigation breakthroughs earlier that benignant of point becomes possible?

From personification who’s building pinch this each day, I deliberation nan capacity is beautiful overmuch there. There are different spectrum complexities of knowledge work. The exemplary is rather smart. I would opportunity what’s lacking is nan plumbing, nan toolings that unlock nan capacity of nan model. That’s fundamentally what Notion is doing, pinch nan LEGO blocks being nan plumbing and tooling. So, that’s 1 constraint. 

The different constraint is conscionable really group usage it, and really group plug it into their workforce. Bureaucracy is sometimes a bully thing, and sometimes it’s a bad thing. In this case, I deliberation it’s really a bully thing, because it slows things down a small bit. It gives group nan clip to accommodate and to study pinch this caller tool. I deliberation it’s good. So, nan capacity is much aliases little there, and if not, each 3 months you get a caller one. The trends conscionable support coming, right?

At nan aforesaid time, I deliberation nan biggest flaw successful nan AI models that we person coming is that they’re not reliable. They don’t reply nan aforesaid mobility nan aforesaid measurement 100 percent of nan time. So, if I’m relying connected it for mission-critical stuff, if it’s 1 of nan 10 “people” astatine my company, and I show it to spell drawback immoderate facts and figures, and it conscionable benignant of hallucinates nan incorrect one, that’s really bad. 

If that were a existent worker, I would, I don’t know, put them connected a capacity betterment scheme aliases something. So really do you deliberation astir reliability arsenic a situation to what kinds of services you want to connection people?

Yeah, it’s decidedly an issue, and I would opportunity it’s getting amended successful general. I would opportunity nan best, closest intelligence exemplary is treating nan connection exemplary conscionable for illustration a human, conscionable for illustration an intern. Humans make mistakes. Your spot level, erstwhile you show different quality something, is that there’s nary guarantee that this quality cannot messiness it up aliases show different person, moreover though you don’t want them to, right?

So, we’re yet building spot this way. People’s expectations for package are higher because software, for nan longest time, was ever nonstop — location are nary bugs. It ever does precisely what it’s told. AI is simply a caller type of software. Our anticipation hasn’t been group connected really to woody pinch this yet. I deliberation arsenic much group get utilized to it, arsenic we alteration our habits astir it and companies alteration their workflows astir it, I deliberation we’ll find an equilibrium that’s amplifying nan amended portion of this exertion and dealing pinch its shortcomings. 

Yeah, I deliberation nan podcaster, Dwarkesh Patel, said thing like, “An AI coming is amended than an intern connected time one, but worse than an intern connected time five.” Because connected time one, they person each nan knowledge of quality history, and they tin benignant of dazzle you pinch their capabilities.

But also, they person problem learning, and it’s difficult to show them really to do thing erstwhile and past person them do it reliably each azygous time. Whereas a quality being could do that. So, I’m personally very curious, erstwhile is nan constituent erstwhile an AI is amended than that day-five intern?

I deliberation each companies, including Notion, are trying to fig retired techniques to inject representation and learning into this “intern.” In nan coming quarters, you will spot products pinch this baked in.

Okay, now my ears are perking up, because it sounds for illustration we’re getting a small spot of a preview. Do you want to show america what you’re moving connected complete there?

Similar to nan run I conscionable talked about. A mates of months ago, we launched Notion AI for Work. It has AI gathering notes and heavy investigation to thief you draught documents. For nan upcoming product, you tin really now ideate each 1 is an AI intern and tin do a specialized thing, right? With nan upcoming product, you tin really create different flavors of AI interns, AI teammates, that unrecorded pinch you successful your workspace. That’s arsenic acold arsenic I want to share. Well, you’ll spot much soon. And Notion AI tin do everything you tin do, everything a quality tin do.

I for illustration nan sound of that. Let maine inquire you astir my merchandise request, and you cognize this request, because I’ve had it for a while now. Basically, erstwhile you first started adding AI features into Notion a fewer years ago, I put successful this request, because each nexus that has ever been successful my newsletter, Platformer, is stored successful a Notion database. In galore cases, that includes nan afloat article text, and what I want is to beryllium capable to person a speech pinch that peculiar database. It would beryllium truthful useful for investigation and brainstorming columns. 

At nan aforesaid time, it’s hundreds of thousands, possibly millions, of words. It’s not nan benignant of point that you could easy propulsion into a discourse model and conscionable fto maine person that conversation. So, my question, Ivan, is wherever are we connected this dream of mine?

To beryllium capable to person a speech pinch each your thousands of articles successful Notion?

Yes.

It is astir apt already there, because nan techniques person been invented. You’re correct that you don’t person to fresh everything into a discourse window. You tin scale everything, embed everything, and portion accusation retired arsenic you request it. There’s different method that’s been celebrated successful nan past twelvemonth aliases so, called instrumentality use. It fundamentally teaches nan connection model, your agent, to cognize really to usage search. 

So, if you person a mobility that’s not successful nan discourse model initially, nan supplier tin spell there, conscionable for illustration a quality can, to find much accusation astir it. It will return a small spot much clip round-trip, but eventually, it will springiness you what you want. So, caller techniques will make nan usage cases you’re describing better.

I for illustration nan sound of that. And you do person an Ask Notion feature already that I ideate tin entree immoderate constituent of what I’m talking about, and a batch of this worldly is conscionable benignant of connected nan web, truthful location are different ways of accessing it. But I conscionable ever think, “Man, if I could person a lightning accelerated measurement of conscionable chatting pinch this database nan measurement I would chat pinch a coworker, that’d beryllium ace cool.”

Oh, it should already beryllium successful your Notion workspace. Happy to locomotion you done nan caller endeavor hunt we conscionable launched. It’s cleanable for that.

Okay, great. All right, we’ll troubleshoot that offline. OpenAI has travel up a mates of times today. You activity intimately pinch nan company. Recently, it announced that you tin usage ChatGPT to create presentations and descent decks. All of nan large labs are moving connected these full-stack virtual assistants that they opportunity mightiness someday beryllium capable to do thing a distant worker mightiness beryllium capable to do. 

Do you deliberation that they’ll get location successful nan next, let’s say, 5 years? And what domiciled do you spot Notion playing successful a world wherever AI’s capabilities are quickly expanding that way?

Yeah, 1 measurement to deliberation astir it is connected nan spectrum, whether it’s much personal, business-to-consumer flavored. Or it’s for illustration business-to-business, team-first flavored.

I would opportunity astir AI labs’ products are presently much individual assistant-flavored. It tin thief you do activity aliases thief you cheat connected your homework. Usually, B2C tends to beryllium winner-takes-all, aliases location are fewer winners, and I deliberation it makes consciousness for labs for illustration OpenAI to spell really difficult successful that direction. Notion squarely is simply a B2B company. Our product, our business model, is for different businesses, and wrong B2B, you’re required to make different tradeoffs. 

There are truthful galore different subcategories, and it has to beryllium team-first to start. That’s why our AI teammates, our AI supplier and AI system, fundamentally unrecorded successful a squad abstraction that you aliases nan remainder of your company’s labor unrecorded in. That’s nan perspective we’re taking. In my opinion, location will beryllium galore different winners successful B2B AI, because B2B usually is not winner-takes-all, and you request to make very different tradeoffs.

That’s interesting. I’m still not judge I wholly understand. Say a spot much astir what this business-to-business AI world looks like, and why it is that it will person galore winners, whereas business-to-consumer possibly doesn’t?

Well, you tin deliberation astir it successful nan master setting. So if you deliberation astir each knowledge work, it is lawyers, accountants, programmers, and customer support, and they’re each different. They require you to make a small spot of a different trade-off; that’s a different type of AI agent. You tin already spot this successful nan first-generation AI agent. They request to beryllium specialized and request to plug into different contexts.

On nan user side, you conscionable want to chat pinch your AI chatbot. It’s very universal. That’s why connected nan B2C side, successful nan erstwhile generation, location were iPhone and Android. There are 2 things, largely, and successful nan B2B SaaS world, location are thousands of different companies and hundreds of different categories. So, whether it’s a package aliases AI product, you person to make very different trade-offs. You cannot beryllium an airplane and a submarine astatine nan aforesaid time, right? That’s why successful nan master setting, you spot a lawyer supplier and a financial agent, and they behave very differently. They request to behave very differently, compared to your individual adjunct that you aftermath up to each time and tin chat with.

Right now, you waste AI devices arsenic an add-on successful your business and endeavor plans. I’m funny if this hurts your separator astatine all? We perceive a batch astir really costly and compute-intensive, resource-intensive AI systems tin beryllium to operate. Is it a situation to merge those resource-hungry devices into your existing subscription?

We really precocious merged AI into our main plan, because much than half of our income are now from customers who want to bargain our AI product. So, it makes consciousness to conscionable simplify nan pricing buckets, to conscionable see AI into everything. It does make nan separator not arsenic bully arsenic axenic SaaS, but no. It’s truthful powerful, and group admit it. And still, nan institution is rate travel positive, truthful our CFO loves that, contempt it being a different separator profile.

We’ve already seen immoderate companies move toward usage-based pricing for AI, which, arsenic a consumer, I hate. I don’t want to make micropayments to inquire ChatGPT a question, but it does look for illustration that is possibly a amended business model. What do you deliberation astir nan tradeoffs there?

I don’t deliberation group person figured it out, particularly successful nan business setting. The first procreation is benignant of for illustration customer support. With customer support, you tin effort to representation what they telephone result resolution-based pricing. That makes sense. Then there’s benignant of nan 2nd generation, which is retired correct now, and that’s coding. With coding, there’s a spot base, but if you usage a lot, you person to spell to usage-based pricing. That benignant of makes sense, because done exchange, it is simply a portion of work. You get your file, you get your package astatine nan extremity of nan day. So group admit that, and it saves programmers truthful overmuch much clip than really penning nan portion of package themselves.

Knowledge activity is nebulous. With knowledge work, you can’t put a value connected it. It’s this chunk of a doc, but really overmuch is it worth? You can’t really quantify that. And really bully is simply a portion of knowledge work? How bully is your merchandise spec? You can’t put a dollar motion connected it. So, it’s overmuch harder for a general-purpose knowledge activity merchandise for illustration Notion. That’s nan thing, nan full manufacture needs to fig it out.

That’s really interesting. What do you wish that AI would make imaginable successful Notion that isn’t rather imaginable yet?

You tin ever get cheaper, faster, and smarter, but you cognize nan train is coming successful that direction, truthful it does require you to build a institution successful a different way. I deliberation this is what nan package manufacture is realizing correct now.

Say much astir that.

I ne'er worked during nan dot com era. That was a small spot earlier me. People opportunity that during that era, nan web modular changed each nan time, each mates of months, aliases each 3 months it’s different. And location was Intel and nan Moore’s Law era, wherever you could conscionable expect that 18 months later, nan adjacent CPU would thrust immoderate package you wanted. 

AI feels for illustration that, but connected steroids. Like each 3 months, nan adjacent exemplary tin do what you couldn’t do before. So it does require you to really alteration really you build package and build products, and really you build a company. A mates of things: One is because it’s perpetually changing, and nan exemplary itself doesn’t for illustration excessively galore restrictions; you request to build a harness conscionable astir nan correct places. It’s almost for illustration if you build excessively overmuch astir nan train track, nan adjacent train comes, and you conscionable made what you conscionable built obsolete, right? You should build parallel to nan train track. That’s number one.

Number 2 is that nan connection exemplary is not deterministic. It’s different from classical package engineering. The metaphor I for illustration to usage is that classical package engineering is for illustration building train tracks aliases building bridges. It’s Newtonian physics. Everything’s predictable, and if you tin ideate it, you tin build it. Sometimes it takes 3 months, sometimes six months, but yet you tin build it, right?

With this connection exemplary thing, it’s squishy and it’s organic. The affinity I emotion to usage is for illustration brewing beer, right? You cannot show nan yeast, “Hey, my brew is gonna sensation for illustration this. Please ferment yourself. Become for illustration that.” You person to transmission what’s successful nan model. The champion you tin do is create an environment, massage nan data, massage nan context, and past dream for nan best. 

So, this requires a overmuch much iterative approach. You cannot travel from your imagination aliases customer needs first. You person to travel from what nan exertion gives you — what nan yeast, what nan beer, gives you. So, really let your squad to beryllium much empirical, much experimental, little of this benignant of waterfall, classical measurement of specing to code. It should beryllium much for illustration incremental, iterative. All those adhd together and unit you to design, engineer, and create products differently. 

Does it alteration nan measurement that you hire? Does it alteration nan measurement that you building teams? How does that strangeness that you picture construe into a different company?

People request to beryllium much okay pinch ambiguity. People request to emotion ambiguity. People request to beryllium much experimental. The bound betwixt roles is going to beryllium moreover greater. Like astatine Notion, we’d prosecute a designer who tin code, because if you’re an technologist and designer successful one, you tin deliberation a batch much ambiguously, much fluidly, right? 

AI clip pushed that moreover further, because nan creation and merchandise beryllium broadside by broadside pinch engineers. Oftentimes, what you want cannot beryllium built, truthful you person to really effort a bunch of different things. That’s why you spot a batch of merchandise demos get to for illustration 60 to 70 percent, but ne'er go a existent product. That’s because it’s bully for making demos, but to get to accumulation B2B software, you request to beryllium really good. You request to beryllium reliable. Oftentimes, you ne'er get there.

I deliberation astir this a batch successful nan discourse of these voice-based assistants. What I mostly usage them for is mounting a timer aliases asking what nan upwind is, these very deterministic things. And nan companies that are building [voice-based assistants] are trying to merge these caller AI-based backends. But it’s incredibly hard, because if nan personification is still utilizing nan product, they’re still going to want to group nan timer, and if it goes from doing it correctly 100 percent of nan clip to for illustration 93 percent of nan time, that’s a overmuch worse product.

But I deliberation arsenic humans, we each study what this type of exertion is champion at. It’s erstwhile you’re having a conversation, for illustration erstwhile utilizing nan sound mode, that you want it to beryllium ambiguous. You want it to spell to different places. That’s a feature, not a bug. I deliberation we — arsenic a full manufacture making package pinch AI, and we arsenic an assemblage who usage it — haven’t figured retired nan stance yet. It will return immoderate clip to fig this out, for illustration nan champion worldly to use.

Well, I want to extremity conscionable by asking what you deliberation Notion looks for illustration a small spot into nan future. I will not inquire you astir 5 years from now, because I don’t deliberation anybody has 5 years’ worthy of visibility into anything.

Nobody knows. 

But if I could possibly inquire for for illustration 2 years from now, what do you dream Notion is doing that it’s possibly not doing today?

I think, to spell backmost to what we conscionable talked about, that nan quality of package is changing. It’s changing and evolving from conscionable a group of devices to this integrated matter, to a instrumentality that tin do immoderate activity for you, right? The bosom of this institution is successful SaaS software. The classical package era allows group to build tools, to let group to usage LEGO to create immoderate devices they want. 

Because nan quality of that package is changing, what we attraction astir allows you to create AI teammates to thief you return immoderate of nan astir repetitive knowledge activity you don’t for illustration to do. If we tin recognize that, location are a batch of implications. The adjacent procreation of builders is going to tally a institution very otherwise — and I attraction astir solving that problem.

All right, well, Ivan, acknowledgment truthful overmuch for joining maine today.

Thank you for having me.

Questions aliases comments astir this episode? Hit america up astatine decoder@theverge.com. We really do publication each email!