Latest news with #CaseyNewton
The Verge
a day ago
- Business
- The Verge
Notion CEO Ivan Zhao wants you to demand better from your tools
Hello, and welcome to Decoder! This is Casey Newton, founder and editor of Platformer and cohost of the Hard Fork podcast. This is the second episode of my productivity-focused Decoder series that I'm doing while Nilay is out on parental leave. Today, I'm talking with Notion cofounder and CEO Ivan Zhao. I've followed Notion for quite some time now — I'm a big fan, and a major part of my workflow for Platformer is actually built on top of Notion's database feature. So I was very excited to get Ivan on the show to discuss his philosophy on productivity, how he's grown his company over the last decade, and where he sees the space going in the future. If you've never used Notion, you can think of it as an all-in-one productivity suite comparable to a lot of the collaboration and so-called 'second brain' apps on the market — from the more business-y project management tools like Asana and AirTable to the more DIY note-taking variants like Anytype and Obsidian. Listen to Decoder, a show hosted by The Verge's Nilay Patel about big ideas — and other problems. Subscribe here! Notion sits pretty comfortably in the middle here, since it can do what a lot of these kinds of apps do very well, and all in one package. At the same time, it allows for a pretty substantial amount of customization, which has made it popular both for individual productivity power users and for companies large and small. But Notion started as a very different piece of software, and its evolution over the last 12 years or so has involved a fair amount of trial and error, one major reboot, and a lot of big decisions. In my opinion, what really sets Notion apart from so many of its peers is Ivan's deep passion for design and an almost relentless drive to make products that he sees as useful and aesthetic in equal measure. You'll hear Ivan early on in this conversation reference LEGO: the toy bricks are the central inspiration for Notion, which employs 'blocks' as a metaphor for configurable templates that allow you to use Notion in a pretty diverse set of ways. Think everything from simple notes and lists to complex databases and workflows. But Notion, like so much software these days, is evolving. Now, the company calls itself the 'AI workspace that works for you,' and you'll hear Ivan recount in detail how the launch of OpenAI's GPT-4 proved to be a big turning point for him and for Notion. The company launched an OpenAI-powered AI product much sooner than the competition, even before the launch of ChatGPT, and it's added a host of new AI-powered features in the past few years. Ivan himself is also pretty excited about the capabilities of AI; he said he uses it in his free time to learn about new subjects, and you'll hear talk in depth here about his vision for AI agents that increasingly do more and more work for you inside of apps like Notion. But a common theme with the AI industry right now is the very large gap between what AI can actually do today, and what so many people hope it can do down the road. So I really wanted to ask Ivan how we might get to this future he predicts, how long that will really take, and what productivity and knowledge work look like if AI delivers on some of these lofty promises. Okay: Notion CEO Ivan Zhao. Here we go. This interview has been lightly edited for length and clarity. Ivan Zhao, you are the co-founder and CEO of Notion. Welcome to Decoder. Thanks for having me. So at a high level, describe Notion for us. If listeners haven't used it yet, what is it? What does it do? Well, we make all-in-one productivity software. People use Notion for all kinds of things, from taking notes, collaborating on projects, managing documents, managing their knowledge base, and, most recently, we launched a calendar product and mail product. You use Notion, so you should describe what Notion is. I think you just did a really good job describing it. I do use Notion, which is one of the reasons I wanted to talk to you, because every time I talk to a CEO whose products I use, I get to give them product feedback, which is exciting for me. Perfect. So do you see Notion today as more for teams than individuals? Is that the direction it has found? We designed Notion for teams. So another way to describe Notion is that we call it LEGO for software. Maybe it's worth explaining the intent a little bit. If you're a company, for your team, you have to use a dozen different tools to get your work done, and our goal is to consolidate those tools into one box and give you the LEGO blocks that power all those use cases. Not only can you do all your work in one place, but you can also use those LEGO blocks to create and customize your own workflows. You and I have talked a fair amount about LEGO over the years. What appeals to you about that design? Why is it such a good metaphor for what you're trying to do with Notion? Well, because it didn't quite exist with software. If you think about the last 15 years of [software-as-a-service], it's largely people building vertical point solutions. For each buyer, for each point, that solution sort of makes sense. The way we describe it is that it's like a hard plastic solution for your problem, but once you have 20 different hard plastic solutions, they sort of don't fit well together. You cannot tinker with them. As an end user, you have to jump between half a dozen of them each day. That's not quite right, and we're also inspired by the early computing pioneers who in the '60s and '70s thought that computing should be more LEGO-like rather than like hard plastic. That's what got me started working on Notion a long time ago, when I was reading a computer science paper back in college. You wanted to make tools that would snap together the way that LEGO blocks do. We want to make tools that amplify human creativity. LEGO are creative. LEGO are beautiful, and most software probably is not as much. Say a little bit more about how you were drawn into this world. Were you always somebody who was interested in productivity tools? Or did that come to you later in life? No. I think it's a misunderstanding about Notion. Notion is productivity software, that's what we do as a business, as a product. But the ethos of this is what I just described; it's the LEGO ethos. It's maybe worth describing the history of the computing industry a little bit, because that inspired Notion. Yeah. In the '60s and '70s, '60s, the hippie generation who took acid in San Francisco thought, 'Well holy shit, this room-sized calculator, if you put a monitor in front of it, it can be an interactive thing, it can be a new type of medium that helps you think better, helps you solve problems more collaboratively.' That's why the first generation of personal computing started in the Bay Area. That generation of thinkers and pioneers thought about computing kind of like reading and writing. Like we went to school for multiple years, but not everybody can read and write English or German or whatever language you speak. Writing is a tool. Yes, you can be a poet, you can be writing essays, but it's a very malleable medium. So, they set out to make computing malleable and tinkerable, and everybody could their own software. Then, in the '80s, the Bill Gates and Steve Jobs generation took computers to the mass market and put a computer in every home, on every desk. They sort of froze computing into this application format. If you think about applications, each application is kind of like a mini-prison of computing. You cannot change it that much. There are application makers who are engineer programmers, and then there are application users who are like the rest of us, the people who use productivity tools every day. So when I was reading those papers about those '60s and '70s people, I thought, 'Holy shit, the world that we're living in is like a prison-like world.' If anything, the SaaS of the past 15 years has been for even smaller prison cells, as each application can only do a tiny slice of things. So, that doesn't make it right for me, and the customer feels the same way. It doesn't make sense that for your daily job, you have to jump between 20 different tools to get some work done. The average company or average business uses 100-plus different SaaS tools. The fragmentation is obvious, even for the IT department. So, there's another saying in business: you either bundle or unbundle. So, Notion's squarely in the bundling business. Our job is to bundle SaaS into a one-ish productivity tool for your core daily needs, so we can unleash LEGO-like creativity for you. It's an interesting conversation. It makes me think of kitchen gadgets, because you see the same tension there, where there are some kitchen tools, like I don't know, a stand mixer, or an immersion blender, that you can use to make many, many different kinds of recipes. And then there's the garlic press, which is good for mincing garlic and nothing else. It sounds like what you're saying is by the time we got into the 2010s, productivity was just a bunch of garlic presses, and you sort of wanted to come along and say, 'What if we just had a stand mixer and you could make a lot of recipes with one thing?' A friend of mine used this metaphor, similar to what you said. Have you seen avocado cutters? Yes. An avocado cutter is made just for freaking avocados. You cannot do anything else. In comparison, a kitchen knife is a tool that you can use hundreds or thousands of different ways. You, as a human, amplify it because you have a technique. So, to create software that's more like a kitchen knife, or LEGO, that's what interests me, and interests us as a company. But you can't blame the industry, because if you think about it — if you rewind back before SaaS — the world was running on Microsoft Office for a good solid 10 to 20 years. SaaS, with the internet as its distribution, allows new businesses to be created. The natural way to go about distribution and new businesses is to find a really precise solution, creating those avocado cutters and garlic presses, and you can find buyers on the other side. So, as we move into today, do you think of yourself as competing directly against Microsoft Office or Google Workplace? Is the vision that big, or is it something different? We coexist with them, as most of our customers are still using Google Workplace or Microsoft Office. They use their identity service, they use their mail and calendar apps. We have a mail and calendar product currently as a client. A startup can fully run on Notion. You don't need to use Microsoft 360 or Google Docs, but it's not as mutually exclusive. Our sweet spot is more on the things that you need to put in a database. Another way to think about it is like, what is Microsoft Access but for the 2020s, and AI native? Most SaaS is kind of like a relational database, storing some kind of system record of your company, and one workflow on top of that. That's the part that neither Microsoft nor Google touches today. There are spreadsheets, but there are not many database use cases. We want to consolidate and commoditize that, and give people the LEGO of those database use cases, such as project management and ticket tracking. Some companies use them for CRM, or managing application trackers. For reporters, you can manage all your leads and the stories. Those are database use cases. Right, and I do do some of that in Notion. Let me ask you about the flip side of building a product that has so much utility baked into it, which is that sometimes when I've talked to people who have tried Notion, they say, 'I didn't know where to begin. I felt intimidated by the blank page.' It seemed like there was a learning curve. How do you think about that challenge, and try to bring people along into understanding what Notion is meant to do? Yeah, like early LEGO, you get bricks. Then later on, LEGO created systems and boxes, and now LEGO works with Marvel and F1 to create really specialized boxes. In some sense, Notion as a company, we're in the middle of adding more boxes, so people don't have to start from an empty set of bricks, with no instruction manual. You can imagine, like, 'Hey, I want a Formula One race car. I like that LEGO box.' When you open it, you have your car ready-made for you, so you can start driving it. You play with that LEGO toy right away. But if you don't like certain parts of the cars, because they're made from LEGOs, you can change them. That has always been our philosophy, and we're doubling down on this approach because it works. Interesting. I feel like a key challenge that some of the other big productivity tools like Microsoft Office have had over the years is bloat, right? The app has a million features, and each individual one is very important to like 0.5 percent of the user base, so you can't remove it. But also, the app just gets harder and harder to use over time because it's so stuffed with buttons, menus, and widgets. Can Notion avoid that? And have there been times when you've worried you might be there already? It's definitely tricky. If you want to support more power, you need to have more things. There are two ways to approach it. The classic way is just adding that feature, in the hard plastic way. We're taking a more LEGO approach, so adding the brick, and the brick can be used for different things. In some sense, this is a much better approach. Got it. So, trying to offer fewer discreet, very narrow features, and more abstract features that can be extended in various different ways. Like LEGO systems, on one end can be toy cars, and on the other could be Barbie dolls, more or less using the same bricks. If you look at the most common productivity tool, if you just put your designer mind on that, there are 20 or 30 pieces there. There's some kind of table, some kind of relational database feature, some charts, some commenting, page editing, and collaboration. Those 20 things are core to all collaboration and knowledge work. So, we try to do our best job to make them friendly, approachable, and break them into pieces, and give them to you, either as a piece or as a part of a package. What bricks inside of Notion are most popular today? We start with bricks around documents and knowledge bases. We're famous for our block-based editors, and that's from the early days. That's like 2019 into 2020. And then databases are our most important brick today, because, like I mentioned, most knowledge work is just fancy file cabinets in the cloud. Knowledge work runs on file cabinets, and databases are the heart of that. Yeah, the database is my number one Notion brick that I use. So that makes sense to me. People don't discover that [easily]. We need to do a better job of getting people to understand its power. It's essentially what an engineer does every day is wire together a relational database with views on top of that. How do we democratize that? That's our purpose. Well, it's interesting, because if I had never heard of Notion, and you came to me and you said, 'Casey, you should build a database to solve this problem,' that's like telling me that I should add another room to my house. I don't know where to begin. I feel like I need to call somebody and ask for help. But in practice, you click a couple of buttons, and in my case, install the Notion Web Clipper, and I'm well on my way to having a database. So, I don't actually think the learning curve is that steep, but I could understand why somebody might be intimidated by it if they had never tried to do that sort of thing. Yeah, not every kid grew up liking LEGO as their number one toy. I think Notion resonates the best with people who like to build, and they tend to be entrepreneurs, tech people, and the spreadsheet gurus in each team. They like Notion, and they set it up for the rest of their teammates. That's always helpful when you can get people inside the company doing the sales part for you. You know what? AI can do this quite well now, because what AI is good at is gluing together LEGO bricks. AI can write code. Writing code is just another way of gluing together your process and workflows, and our latest product essentially gets AI to be this successful person to help set up your Notion workspace for you, and that's another way to onboard customers. That's a brand new way to unlock that we've added in the past year to two years. I will say that it has been very powerful for me in a lot of different products. Being able to use the in-app AI to say, 'How do I do this?' And actually getting an answer. As somebody who has spent a lot of time in help menus over the years, digging around and not finding what I was looking for, that has been super useful. And it's not just helping by teaching the human to do it. More and more, AI can just do it, right? Right. That's the biggest difference, actually. If you think about what's happening in software right now, software is largely people providing the tools for humans to use, and more and more companies are realizing, 'Wait a second, we have this new thing called a language model. It's like a human mini-intern in a box, and we should design our software to teach AI how to use it so humans can ask AI to do the work and use the tools, and humans can do way more things with it.' I want to ask you some of the Decoder questions that Nilay would ask if he were here. Notion was last valued at $10 billion nearly four years ago, when you raised your last round of funding. What has allowed you to keep growing without raising more funding? Are you profitable? We're profitable. So, profitable, growing fast; the business is doing well. Nice. How does that feel? It feels good. I would say the larger driver of our everyday activity is the fact that the software industry is completely changing with AI. It feels like the AI era of the past two years just makes the SaaS era feel like sleeping days. It's a bigger driver of our execution strategy than just running a profitable business. I'd like to hear more about how that is working. Is it the case that executives see AI changing various workplaces, and they think, 'We need to figure out our version of this,' and so they come to Notion to help them figure it out? Or is it that your product teams are so excited about the possibilities that you're just now seeing them build features, which are then drawing in new customers? I would say customers are lagging behind at the moment. Most people don't know. It happens with every new technology. You don't know what to do with it. The customer is not going to tell you. It's the people who play with this, build things, and maybe have an imagination a few years into the future, or even a few months in the future, at this point. AI is changing so fast. So a lot is from ourselves, just playing with AI and realizing, 'Holy moly, this is a very different thing. You can solve problems that you couldn't solve before with classic software.' Now, what are you going to do with it? There's actually an interesting story. My co-founder, Simon Last, and I received early access to GPT-4. So this is like late 2022, a little bit before everybody else. We thought everybody else would get early access to this. We thought, 'Holy shit,' because compared to GPT-3, GPT-4 is a brand-new thing. It's like it has real intelligent reasoning in it. So we locked ourselves in a hotel room for about a week, and just tried to rush out the first Notion AI product. We actually launched a month before ChatGPT happened. We were excited about what you can do with this new type of material. That's for Notion, that energy comes from there. How many employees do you have over there? How big is Notion today? High three digits. So, 900, maybe approaching 1,000. How do you think about company size? Do you see a world where there are five times that many employees? Or do you want to keep it somewhere around where it is right now? I think there's no right answer for the absolute number, but there's an answer for the density of the talent. The denser the better. You like having fewer, but more talented people as opposed to more people? I like to have fewer, if we can do the work with fewer people, and there's less communication overhead. People have more ownership, and people can work things across boundaries. That's just better overall. The company moves faster. The small car can turn corners much better than a big car, and we always call Notion a small bus. We try to keep the bus as tight as possible. What's your org chart? How is Notion organized? Fairly classic. It's me and my co-founders, Simon Last and Akshay Kothari. Simon is still coding every day. Akshay runs our product and design org and research. Our chief technology officer, Fuzzy Khosrowshahi, runs all of the engineering. Our chief revenue officer, Erica Anderson, is responsible for sales, marketing, and consumer experience. And we have our chief finance officer, Rama Katkar, and general counsel, Hasani Caraway. That's our classic org chart. So you didn't feel the need to reinvent the wheel there or do any innovating, just sort of create classic company divisions and let people go do their thing. Classic company divisions and high-quality people keep the bus tight. So that allows you to be profitable. How do you make big decisions? Do you have a framework you use? Or is every decision different? Well, there are the typical one-way doors and two-way doors. With a one-way door, you try to move fast, and with a two-way door, you think a little bit more carefully and sleep on it. Those are thinking fast and slow types of things. I'm pretty detailed. I like to work on the notes, so there's a certain problem I'm good at. Personally, it gives me energy, and I'm interested in working on the ground in the trenches with everybody. There are certain parts, though, like I cannot run our finance team. Our CFO, Rama Katkar, is really amazing. She takes care of that. But for certain things — like design and product, engineering, marketing, and branding — I like to get involved. You've always struck me as a product CEO. I think from the first time I met you, it seemed to me what was most interesting to you about your company was the tool itself that you were building, as opposed to the market opportunity, or something like that. I built Notion because I wanted to build Notion, not because I wanted to start a company or business. I wanted this thing to exist. Let me ask you one more decision question, about one of the bigger decisions you had to make. So in 2015, you decided to shut down the 1.0 version of Notion, relocate to Japan, and eventually relaunch Notion 2.0, which is kind of the Notion that we think of when we use it today. How did you make that decision? Well, you have to, otherwise you die, right? At that point, it was like, 'We're building on the wrong thing, the wrong foundation,' and you know what the right thing is, but it's just going to take you maybe a year to a year and a half to build the right thing. We were a company of about five people, and we were going to run out of money. The only way to do it was to shrink it back to just me and my co-founder, Simon. So, we started over. Japan is a good place because it is inexpensive, and we have never been there. It was interesting, and we could just focus on building. I know other founders who have been in that situation, and that's the moment where they gave up, because they thought, 'You know what? Maybe I could think of another thing to build here, but it seems exhausting. It's going to take a year. I've already put a year and a half or so 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 keep going on this. We think that there is a vision here that we can actually achieve'? The goal was never to start a business, like I mentioned. The goal was to build this thing, and the thing didn't quite exist. Notion is one of the few bundling, consolidating software productivity tools out there, and it didn't quite exist at the time. Software for LEGO doesn't exist. So if I started a company, I wanted to do the same thing. Why don't I just reset and go back to Simon and me so we can stretch the money a bit longer? Actually, I just got back from Kyoto last week, where there was a tech event, and the Kyoto mayor and I did a fireside chat and talked about this story. They wanted to talk about using Notion as an example of how we can blend tech and Kyoto's craft tradition. Because we're also inspired a lot by the craft people in Kyoto, how they dedicate their time to building something, and not just for money or fame. That is so crazy. I was actually in Kyoto last week, too. I was on vacation and went there for the first time, and I had an incredible time. Kyoto's amazing. You get the sense that it's a little bit slower pace. People care about what [they're making], right? People truly care. That's the thing. It's the main thing. It's not the business, it's not their other surroundings. Oh absolutely. I mean, you go to these temples that are 1,000-plus years old, and the care and the craftsmanship that they put into them is truly inspiring. It's deeply beautiful. It's very connected to their spirituality, and their religion, culture, and history. So I could understand why a founder would go there and take a lot of inspiration. Yeah, because it's a bit slower too, you can focus on virtual spaces, on computers. Yeah, it's not like San Francisco, with our go-go nightlife, our Waymos, the party scene, all of that. Or New York, even, even more of that. One more of these. This isn't strictly a Decoder question, but it spiritually feels like a Decoder question that I wanted to ask you. What is the best productivity tool that you use that is not made by Notion? I like those chatbot products: ChatGPT, Anthropic's Claude. It's quite amazing, especially when talking about features, like the conversation mode. I love those. I like to learn from them. You like voice mode? The voice mode, yeah. It helps me learn a lot of different things. When I'm making coffee and waiting for the water to boil, I can just talk with this thing for a little bit, like for two minutes. It's perfect. What do you ask it about? What do you like to learn about? Oh, all kinds. What's most recent? In Japan, I was reading a book about Marshall McLuhan. The media theorist? Yeah, media theorist and theologian. Many of his concepts are hard to interpret, so it's better to just work through them with a language model that will help guide you. It's the best tutor, truly. Education should be very different. Hopefully, it will be very different a few years from now. I think so. This isn't quite a tutoring use case, but I just have to say, when I was in Kyoto last week, we were in the neighborhood and had some time to kill, so I just opened Google Maps, and it sort of opens to where we are, and I took a screenshot. I just sent it to ChatGPT, and I said, 'Tell us a bit about this neighborhood.' It gave me a history of the neighborhood. It told me about the cool restaurants, cool coffee shops, a museum, and places that we could walk to. I mean, it truly was as good as I can imagine getting from any guide, and it was as simple as uploading a screenshot. The whole thing took 15 seconds. It was wild to me. Yeah. One use case I have, if I go to a famous architectural building, is that I say, 'Hey, I'm at this place, tell me about it. I'm looking at this part, a corner of this building. Tell me more about why that is the case,' right? Because if it's famous enough, it's probably part of the corpus of training. So, the language model knows about it. Yeah. You can just take a truly guided tour, and you don't need another person; you're just talking with your machine. Well, that seems like a good segue into Notion and AI. We've talked about what Notion is, how it's changed. Notion now bills itself as the 'AI workspace that works for you.' So, what does an AI workspace mean to you? What do you want it to be for us? If you think about our strategy during the SaaS era, it's bundling and consolidating a bunch of different tools for knowledge work into one place. What's changed in the past couple of years is that it now has all that soft knowledge of LEGO in Notion; you can not only provide the tools, but you can also assemble them as your AI teammates. They can do the work for you. We are fortunate to have that knowledge work LEGO in one place so you can piece it together in a very interesting way. Because one end can take notes for you, the other end can help you manage, triage projects, and write documents. Those are basic things, but more and more, with more LEGO and smarter models, you essentially hire Notion as your AI teammate. That's the future that we've been building, or building more toward. I remember one time I was meeting with you, and you just launched some of these AI tools, and you were showing me that Notion AI was taking notes about various meetings. So you were able to dip into meetings at the company that you did not personally attend, and just kind of quickly catch up on what your coworkers were talking about. I thought, 'That's super interesting.' That's the kind of feature that I can imagine a lot of CEOs wanting, but before this point, they haven't had that level of visibility into their own company. We just launched three separate products a couple of months ago, like Notion AI for Work, including the next version of this enterprise search product you're talking about. So, along with that was the launch of our AI Meeting Notes product, so all your meetings can be recorded and transcribed. So essentially, your company has a collective brain of what's going on, and you have all the AI knowledge workers on top of Notion to help you transcribe the meeting notes and answer whatever questions you may have. It's quite interesting what you can do with the technology now. Of all the AI features that you've added so far, which ones do you personally find the most useful? I use AI meeting notes. Almost every meeting, except this one, I record, and I use that for meetings so I can share notes with other people. I use it for myself, as a starting position to dump my thoughts. And so I can later remember to ask AI to turn the transcription into writing. English is my second language, so I'm not the fastest writer, but AI can do better writing than I can if I just dump out what's on the top of my mind through the AI transcription feature. That's really interesting. There's a lot of talk right now about AI, and whether it might replace workers, or entire workflows, or functions within an organization. You've talked today about AI being able to serve as a kind of teammate. Do you think that AI and Notion will get to a point where executives will hire fewer people, because Notion will do it for them? Or are you more focused on just helping people do their existing jobs? We're actually putting out a campaign about this, in the coming weeks or months. We want to push out a more amplifying, positive message about what Notion can do for you. So, imagine the billboard we're putting out. It's you in the center. Then, with a tool like Notion or other AI tools, you can have AI teammates. Imagine that you and I start a company. We're two co-founders, we sign up for Notion, and all of a sudden, we're supplemented by other AI teammates, some taking notes for us, some triaging, some doing research while we're sleeping. So, all of a sudden, we're a team or a company of 10 people. Then the startup can run much faster. That's the vision we want to push more towards the world. So, more of an amplifying force, rather than a zero-sum force. What timeframe do you think that arrives? Does that feel like it's almost within reach? Or do you think we're going to need to see several more research breakthroughs before that sort of thing becomes possible? From someone who's building with this every day, I think the capability is pretty much there. There are different spectrum complexities of knowledge work. The model is quite smart. I would say what's lacking is the plumbing, the toolings that unlock the capability of the model. That's essentially what Notion is doing, with the LEGO blocks being the plumbing and tooling. So, that's one constraint. The other constraint is just how people use it, and how people plug it into their workforce. Bureaucracy is sometimes a good thing, and sometimes it's a bad thing. In this case, I think it's actually a good thing, because it slows things down a little bit. It gives people the time to adapt and to learn with this new tool. I think it's good. So, the capability is more or less there, and if not, every three months you get a new one. The trends just keep coming, right? At the same time, I think the biggest flaw in the AI models that we have today is that they're not reliable. They don't answer the same question the same way 100 percent of the time. So, if I'm relying on it for mission-critical stuff, if it's one of the 10 'people' at my company, and I tell it to go grab some facts and figures, and it just kind of hallucinates the wrong one, that's really bad. If that were a real worker, I would, I don't know, put them on a performance improvement plan or something. So how do you think about reliability as a challenge to what kinds of services you want to offer people? Yeah, it's definitely an issue, and I would say it's getting better in general. I would say the best, closest mental model is treating the language model just like a human, just like an intern. Humans make mistakes. Your trust level, when you tell another human something, is that there's no guarantee that this human cannot mess it up or tell another person, even though you don't want them to, right? So, we're finally building trust this way. People's expectations for software are higher because software, for the longest time, was always exact — there are no bugs. It always does exactly what it's told. AI is a new type of software. Our expectation hasn't been set on how to deal with this yet. I think as more people get used to it, as we change our habits around it and companies change their workflows around it, I think we'll find an equilibrium that's amplifying the better part of this technology and dealing with its shortcomings. Yeah, I think the podcaster, Dwarkesh Patel, said something like, 'An AI today is better than an intern on day one, but worse than an intern on day five.' Because on day one, they have all the knowledge of human history, and they can sort of dazzle you with their capabilities. But also, they have trouble learning, and it's hard to show them how to do something once and then have them do it reliably every single time. Whereas a human being could do that. So, I'm personally very curious, when is the point when an AI is better than that day-five intern? I think all companies, including Notion, are trying to figure out techniques to inject memory and learning into this 'intern.' In the coming quarters, you will see products with this baked in. Okay, now my ears are perking up, because it sounds like we're getting a little bit of a preview. Do you want to tell us what you're working on over there? Similar to the campaign I just talked about. A couple of months ago, we launched Notion AI for Work. It has AI meeting notes and deep research to help you draft documents. For the upcoming product, you can actually now imagine each one is an AI intern and can do a specialized thing, right? With the upcoming product, you can actually create different flavors of AI interns, AI teammates, that live with you in your workspace. That's as far as I want to share. Well, you'll see more soon. And Notion AI can do everything you can do, everything a human can do. I like the sound of that. Let me ask you about my product request, and you know this request, because I've had it for a while now. Basically, when you first started adding AI features into Notion a few years ago, I put in this request, because every link that has ever been in my newsletter, Platformer, is stored in a Notion database. In many cases, that includes the full article text, and what I want is to be able to have a conversation with that particular database. It would be so useful for research and brainstorming columns. At the same time, it's hundreds of thousands, maybe millions, of words. It's not the sort of thing that you could easily throw into a context window and just let me have that conversation. So, my question, Ivan, is where are we on this dream of mine? To be able to have a conversation with all your thousands of articles in Notion? Yes. It is probably already there, because the techniques have been invented. You're correct that you don't have to fit everything into a context window. You can index everything, embed everything, and piece information out as you need it. There's another technique that's been popular in the past year or so, called tool use. It essentially teaches the language model, your agent, to know how to use search. So, if you have a question that's not in the context window initially, the agent can go there, just like a human can, to find more information about it. It will take a little bit more time round-trip, but eventually, it will give you what you want. So, new techniques will make the use cases you're describing better. I like the sound of that. And you do have an Ask Notion feature already that I imagine can access some element of what I'm talking about, and a lot of this stuff is just sort of on the web, so there are other ways of accessing it. But I just always think, 'Man, if I could have a lightning fast way of just chatting with this database the way I would chat with a coworker, that'd be super cool.' Oh, it should already be in your Notion workspace. Happy to walk you through the new enterprise search we just launched. It's perfect for that. Okay, great. All right, we'll troubleshoot that offline. OpenAI has come up a couple of times today. You work closely with the company. Recently, it announced that you can use ChatGPT to create presentations and slide decks. All of the big labs are working on these full-stack virtual assistants that they say might someday be able to do anything a remote worker might be able to do. Do you think that they'll get there in the next, let's say, five years? And what role do you see Notion playing in a world where AI's capabilities are rapidly expanding that way? Yeah, one way to think about it is on the spectrum, whether it's more personal, business-to-consumer flavored. Or it's like business-to-business, team-first flavored. I would say most AI labs' products are currently more personal assistant-flavored. It can help you do work or help you cheat on your homework. Usually, B2C tends to be winner-takes-all, or there are few winners, and I think it makes sense for labs like OpenAI to go really hard in that direction. Notion squarely is a B2B company. Our product, our business model, is for other businesses, and inside B2B, you're required to make different tradeoffs. There are so many different subcategories, and it has to be team-first to start. That's why our AI teammates, our AI agent and AI system, fundamentally live in a team space that you or the rest of your company's employees live in. That's the angle we're taking. In my opinion, there will be many different winners in B2B AI, because B2B usually is not winner-takes-all, and you need to make very different tradeoffs. That's interesting. I'm still not sure I totally understand. Say a bit more about what this business-to-business AI world looks like, and why it is that it will have many winners, whereas business-to-consumer maybe doesn't? Well, you can think about it in the professional setting. So if you think about all knowledge work, it is lawyers, accountants, programmers, and customer support, and they're all different. They require you to make a little bit of a different trade-off; that's a different type of AI agent. You can already see this in the first-generation AI agent. They need to be specialized and need to plug into different contexts. On the consumer side, you just want to chat with your AI chatbot. It's very universal. That's why on the B2C side, in the previous generation, there were iPhone and Android. There are two things, largely, and in the B2B SaaS world, there are thousands of different companies and hundreds of different categories. So, whether it's a software or AI product, you have to make very different trade-offs. You cannot be an airplane and a submarine at the same time, right? That's why in the professional setting, you see a lawyer agent and a financial agent, and they behave very differently. They need to behave very differently, compared to your personal assistant that you wake up to every day and can chat with. Right now, you sell AI tools as an add-on in your business and enterprise plans. I'm curious if this hurts your margin at all? We hear a lot about how expensive and compute-intensive, resource-intensive AI systems can be to operate. Is it a challenge to integrate those resource-hungry tools into your existing subscription? We actually recently merged AI into our main plan, because more than half of our sales are now from customers who want to buy our AI product. So, it makes sense to just simplify the pricing buckets, to just include AI into everything. It does make the margin not as good as pure SaaS, but no. It's so powerful, and people appreciate it. And still, the company is cash flow positive, so our CFO loves that, despite it being a different margin profile. We've already seen some companies move toward usage-based pricing for AI, which, as a consumer, I hate. I don't want to make micropayments to ask ChatGPT a question, but it does seem like that is maybe a better business model. What do you think about the tradeoffs there? I don't think people have figured it out, especially in the business setting. The first generation is kind of like customer support. With customer support, you can try to map what they call outcome resolution-based pricing. That makes sense. Then there's sort of the second generation, which is out right now, and that's coding. With coding, there's a seat base, but if you use a lot, you have to go to usage-based pricing. That sort of makes sense, because through exchange, it is a piece of work. You get your file, you get your software at the end of the day. So people appreciate that, and it saves programmers so much more time than actually writing the piece of software themselves. Knowledge work is nebulous. With knowledge work, you can't put a price on it. It's this chunk of a doc, but how much is it worth? You can't really quantify that. And how good is a piece of knowledge work? How good is your product spec? You can't put a dollar sign on it. So, it's much harder for a general-purpose knowledge work product like Notion. That's the thing, the whole industry needs to figure it out. That's really interesting. What do you wish that AI would make possible in Notion that isn't quite possible yet? You can always get cheaper, faster, and smarter, but you know the train is coming in that direction, so it does require you to build a company in a different way. I think this is what the software industry is realizing right now. Say more about that. I never worked during the dot com era. That was a little bit before me. People say that during that era, the web standard changed all the time, every couple of months, or every three months it's different. And there was Intel and the Moore's Law era, where you could just expect that 18 months later, the next CPU would drive whatever software you wanted. AI feels like that, but on steroids. Like every three months, the next model can do what you couldn't do before. So it does require you to really change how you build software and build products, and how you build a company. A couple of things: One is because it's constantly changing, and the model itself doesn't like too many restrictions; you need to build a harness just around the right places. It's almost like if you build too much around the train track, the next train comes, and you just made what you just built obsolete, right? You should build parallel to the train track. That's number one. Number two is that the language model is not deterministic. It's different from classic software engineering. The metaphor I like to use is that classic software engineering is like building train tracks or building bridges. It's Newtonian physics. Everything's predictable, and if you can imagine it, you can build it. Sometimes it takes three months, sometimes six months, but eventually you can build it, right? With this language model thing, it's squishy and it's organic. The analogy I love to use is like brewing beer, right? You cannot tell the yeast, 'Hey, my beer is gonna taste like this. Please ferment yourself. Become like that.' You have to channel what's in the model. The best you can do is create an environment, massage the data, massage the context, and then hope for the best. So, this requires a much more iterative approach. You cannot come from your vision or customer needs first. You have to come from what the technology gives you — what the yeast, what the beer, gives you. So, really allow your team to be more empirical, more experimental, less of this kind of waterfall, classic way of specing to code. It should be more like incremental, iterative. All those add together and force you to design, engineer, and develop products differently. Does it change the way that you hire? Does it change the way that you structure teams? How does that strangeness that you describe translate into a different company? People need to be more okay with ambiguity. People need to love ambiguity. People need to be more experimental. The boundary between roles is going to be even greater. Like at Notion, we'd hire a designer who can code, because if you're an engineer and designer in one, you can think a lot more ambiguously, more fluidly, right? AI time pushed that even further, because the design and product sit side by side with engineers. Oftentimes, what you want cannot be built, so you have to really try a bunch of different things. That's why you see a lot of product demos get to like 60 to 70 percent, but never become a real product. That's because it's good for making demos, but to get to production B2B software, you need to be really good. You need to be reliable. Oftentimes, you never get there. I think about this a lot in the context of these voice-based assistants. What I mostly use them for is setting a timer or asking what the weather is, these very deterministic things. And the companies that are building [voice-based assistants] are trying to integrate these new AI-based backends. But it's incredibly hard, because if the user is still using the product, they're still going to want to set the timer, and if it goes from doing it correctly 100 percent of the time to like 93 percent of the time, that's a much worse product. But I think as humans, we all learn what this type of technology is best at. It's when you're having a conversation, like when using the voice mode, that you want it to be ambiguous. You want it to go to different places. That's a feature, not a bug. I think we — as a whole industry making software with AI, and we as an audience who use it — haven't figured out the stance yet. It will take some time to figure this out, like the best material to use. Well, I want to end just by asking what you think Notion looks like a little bit into the future. I will not ask you about five years from now, because I don't think anybody has five years' worth of visibility into anything. Nobody knows. But if I could maybe ask for like two years from now, what do you hope Notion is doing that it's maybe not doing today? I think, to go back to what we just talked about, that the nature of software is changing. It's changing and evolving from just a set of tools to this organic matter, to a tool that can do some work for you, right? The heart of this company is in SaaS software. The classic software era allows people to build tools, to allow people to use LEGO to create whatever tools they want. Because the nature of that software is changing, what we care about allows you to create AI teammates to help you take some of the most repetitive knowledge work you don't like to do. If we can realize that, there are a lot of implications. The next generation of builders is going to run a company very differently — and I care about solving that problem. All right, well, Ivan, thanks so much for joining me today. Thank you for having me. Questions or comments about this episode? Hit us up at [email protected]. We really do read every email! A podcast from The Verge about big ideas and other problems. Posts from this author will be added to your daily email digest and your homepage feed. See All by Casey Newton Posts from this topic will be added to your daily email digest and your homepage feed. See All AI Posts from this topic will be added to your daily email digest and your homepage feed. See All Apps Posts from this topic will be added to your daily email digest and your homepage feed. See All Business Posts from this topic will be added to your daily email digest and your homepage feed. See All Decoder Posts from this topic will be added to your daily email digest and your homepage feed. See All Podcasts Posts from this topic will be added to your daily email digest and your homepage feed. See All Tech
The Verge
04-08-2025
- The Verge
Why tech is racing to adopt AI coding
Hello, and welcome to Decoder! This is Casey Newton, founder and editor of the Platformer newsletter and cohost of the Hard Fork podcast. I'll be guest hosting the next few episodes of Decoder while Nilay is out on parental leave, and I'm very excited for what we have planned. If you've followed my work at all, particularly when I was a reporter at The Verge, you'll know that I'm a total productivity nerd. At their best, productivity apps are the way we turn technological advancement into human progress. And also: they're fun! I like trying new software, and every new tool brings the hope that this will be the one that completes the setup of my dreams. Listen to Decoder, a show hosted by The Verge's Nilay Patel about big ideas — and other problems. Subscribe here! Over the years, I've used a lot of these programs, but I rarely get a chance to talk to the people who make them. So, for my Decoder episodes, I really wanted to talk to the people behind some of the biggest and most interesting companies in productivity about what they're building and how they can help us get things done. That brings me to my guest today: Michael Truell, the CEO of Anysphere. You may not have heard of Anysphere, but you've likely heard the name of its flagship product: Cursor. Cursor is an automated programming platform that integrates with generative AI models from Anthropic, OpenAI, and others to help you write code. Cursor is built into a standard version of what programmers call an integrated development environment, or IDE, with technology like Cursor Tab, which autocompletes lines of code as you write. Cursor has quickly become one of the most popular and fastest-growing AI products in the world, and Anysphere, the company Michael cofounded just three years ago after graduating from the Massachusetts Institute of Technology, is now shaping up to be one of the biggest startup success stories of the post-ChatGPT era. So I sat down with Michael to talk about Cursor, how it works, and why coding with AI has seen such incredible adoption. As you'll hear Michael explain, this entire field has evolved very quickly over the past few years — and here in San Francisco, tech executives and employees regularly tell me about how much their employees love using Cursor. AI critics are worried that this technology could automate jobs, and rightly so — but you'll hear Michael say that job losses won't come from simple advances in tools like the one he's making. And while a lot of people in the Bay Area believe superintelligent AI is going to remake the world overnight, making products like Cursor pointless, Michael actually believes change is going to come much more slowly. I also wanted to ask Michael about the phenomenon of vibe coding, which lets amateurs use tools like Cursor to experiment in building software of their own. That's not Cursor's primary audience, Michael tells me. But it is part of this broader shift in programming, and he's convinced that we're only just scratching the surface of how much AI can really do here. Okay: Anysphere CEO Michael Truell. Here we go. This interview has been lightly edited for length and Truell, you are the cofounder and CEO of Anysphere, the parent company of Cursor. Welcome to Decoder. Thank you for having me. So what is Cursor? What does it do, and who is it for? Our intention with Cursor is to have it be the best way to build software and, specifically, the best way to code with AI. For people who are nontechnical, I think the best way to think about Cursor, as it exists today, is as a really souped-up word processor in which engineers build software by actually doing a lot of writing. They're sitting in something that looks like a word processor, and they're editing millions of lines of logic — things that don't look like language. Cursor helps them do that work way more efficiently, especially with AI. There's two different ways Cursor does this right now. One is that as Cursor watches you do your work, it tries to predict the next set of things you're going to do within Cursor. So this is the autocomplete form factor, which can be really souped up in programming when compared with writing, because in programming, unlike in writing, oftentimes the next 20 minutes of your work are entirely predictable. Whereas in writing, it can be a little hard to get a sense of what a writer is going to put down on the page. There isn't enough information in the computer to understand the next set of things the writer is going to do. The other way people work with Cursor is by increasingly delegating to it, as if they're working with a pair programmer, another human. They're handing off small tasks to Cursor and having Cursor tend to them. Well, we'll dig a little deeper into the product in a moment. But first let's talk about how all of this started. When you founded Anysphere, you were working on computer-aided design (CAD) software. How did you get from there to Cursor? My cofounders and I had been programming for a while, and we'd also been working on AI for almost as long as we'd been programming. One of my cofounders had worked on recommendation systems in Big Tech. Another had worked on computer-vision research for a long time, while another had worked on trying to make machine learning algorithms that could learn from very, very, very little data. One of us had even worked on a competitor to Google, using the antecedents that came before LLM technology in machine learning. But we'd worked on AI for a long time and had also been engineers for a long time and loved programming. In 2021, there were two moments that really excited us. One was using some of the first really useful AI products. Another was this body of literature that showed that AI was going to get better, even if we ran out of ideas, by making the models bigger and training them on more data. That got us really excited about a formula for creating a company, which was to pick an area of knowledge work and build the best product for that area of knowledge work — a place where you do your work as AI starts to change. And then, the hope is that you do that job well, and you get lots of people to use your product and you can see where AI is helping them and where AI is not helping them — and where the human just has to correct AI a bunch or do the work without any AI help. You can use that to then make the product better and push the underlying machine- learning technology forward. That can maybe get you onto a path where you can really start to build the future of knowledge work as this technology gets more mature, and be the one to push the underlying tech too. So, we got kind of interested in that formula for making a company, but the craft that we really loved, the knowledge work that we really loved, was building things on computers, and we actually didn't touch that at first. We went and we worked on a different area, which was, as you noted, computer-aided design. We were trying to help mechanical engineers, which was a very ill-fitted decision, because none of the four of us were mechanical engineers. We had friends who were interested in the area. We had worked on robotics in the past, but it wasn't really our specialty. We did it because it seemed there were a bunch of other people working on trying to help programmers become more productive as AI got better. But after six or so months of working on the mechanical engineering side of things, we got pulled back into working on programming, and part of that was just our love for the space. Part of it, too, was that it seemed as if the people who we thought had the space covered were building useful things, but they weren't pointed in the same direction and they didn't really seem to be approaching the space with the requisite ambition. So we decided to build the best way to code with AI, and that's where Cursor started. I have read that one of the AI tools that you used early on was GitHub Copilot, which came out about a year before ChatGPT. What was your initial reaction to Copilot, and how did it influence what you wanted to build? Copilot was awesome. Copilot was a really, really big influence, and it was the first product that we used that had AI really at its core that we found useful. One of the sad things to us as people who had been working on AI and interested in AI for a while was that it was very much stuff that was just in the lab or in the toy stage. It felt like, for us, the only real way AI had touched our lives as consumers was mostly through recommendation systems, right? The news feeds of the world, YouTube algorithms, and things like that. GitHub Copilot was the first product where AI was really, really useful at its core and that wasn't vaporware. So, Copilot was a big inspiration, and at the time we were considering whether we should try to pursue careers in academia. Copilot was proof that no, it was time to work on these systems out in the real world. Even back then, in 2021, there were some rough edges. There were some places where the product was wrong in really obvious ways, and you couldn't completely trust its code output, but it was nonetheless really, really exciting. Another thing to note is that apart from being the first useful AI product, Copilot was the most useful new development tool that we had adopted in a really long time. We were people who had optimized our setups as programmers and modded out our text editors and other things like that. We were using this crazy kind of text editor called Vim at the time. So, it was not only the first useful AI product that we had used, but also the most useful dev flow we had used in a really long time. That's interesting. So you all like software, you like using software, you're trying to find software that makes you more productive. I feel like that probably made you well-suited to tackle a problem, the one Cursor is trying to solve. Yeah, I think caring about the tools we use was helpful, and I think that there were actually different degrees of that on the cofounding team. One of my cofounders is straight out of central casting, an early adopter who is the first one on these new browsers, first one on the new category of everything. A couple of us are a little bit more laggard, and so I think having that diversity of opinions has helped us in some of the product decisions we've made. So you described Cursor as kind of like a souped-up word processor. Software engineers I think would call it an integrated development environment, or an IDE. Developers have been using IDEs since the '80s, but recently, AI labs have released tools, like OpenAI's Codex or Anthropic's Claude Code, that can run directly in a terminal. Why might someone use Cursor over those options? I think that both of those are really useful tools. What we care about being, I think we start as this IDE, as this text editor, but what we really care about is getting to a world where programming has completely changed, in particular a world where you can develop professional-grade software, perhaps without even really looking at the code. And, yeah, it's that kind of future programming and changing it from this weird, you're reading these millions of lines of logic and these esoteric programming languages. The world we want to get to is one where you just need to testify the minimal intent necessary to build the software you want. You can tell the computer the shortest amount of information it needs to really get you, and it can fill in all of the gaps. Programming today is this intensely labor-intensive, time-intensive thing, where to do things that are pretty simple to describe, to get them to actually work and show up on a computer, takes many thousands of hours and really large teams and lots of work, especially at professional scale. So that's where we want to get to — inventing that new form of programming. I think that that starts as an editor and then that starts to evolve. So we're already in the midst of that. Right now, Cursor is where you can work one-on-one with an agent, and with our Tab system. And then, increasingly, we're getting you to a world where more and more of programming is moving toward delegating your work to a bunch of helpers in parallel. And there's a product experience to be built for making that great and productive, with an understanding of what all of these parallel helpers are doing for you — diving in, intervening in places where it's helpful, understanding their work when they come back to you at a level of not having to read every single line of code. I think that there's a competitive environment with a bunch of tools that are interested in programming productivity. One of the things that's limiting about just a terminal user interface is that you have only so much expressiveness in the terminal and control over the UI. From the very start, we've thought that the solution to automating code and replacing it with something better is this kind of two-pronged approach, where you need to build the pane of glass where programmers do their work, and you need to discover what the work looks like. You need to build the UI, and then you also need to build the underlying technology. So, one thing that would distinguish us between some terminal tools is just the degree of control you have over the UI. We've also done a lot of work on the model layer, on improving it and going beyond just having things that show up well on a demo level. There's a lot of work on AI products to dial in the speed and the robustness and the accuracy of them. For us, one important product lever has been building an ensemble of models that work with the API models to improve their abilities. So, every time you call out to an agent in Cursor, it's like this set of models — some of them are APIs, some of them are custom — and then for some form factor or for some of the features, it's entirely custom, like for the super autocomplete. That's also one thing that has kind of distinguished us from other solutions. Let's talk a bit about these proprietary models. They seem to be fueling a lot of your success. When ChatGPT and the OpenAI API first got released, we saw a lot of startups come out that were quickly dismissed as just wrappers for an API that was just trying to build something on top of somebody else's tech. Cursor started in a similar way in that it was using other folks' APIs in order to create its product. Since then, you've started to build on top. Say a bit more about what you're building and how you're hoping it sets you apart from those pure wrapper companies. I think also one asterisk before getting into the model side of things is that the 'wrapper' term came from the very start of when people were building AI products, when there was only so much time to make the products a bit deeper. Now, I think we're at a point where there's a ton of product overhang. So even if you're just building with the API models, I think that in a lot of areas — our area of working on the software development lifecycle, but in other parallel areas too — there are very, very deep products to be built on top of those things. So it sounds like the wrapper term for at least some areas is a little bit dated. But on the model level, I think that from the very start we wanted to build a product that got a lot of people using it. One of the benefits you get from that scale is you can see where AI is helping people, and you can see where AI is not helping people and where it gets corrected. That's a really, really important input to making AI more useful for people. So at this point our Tab model, which does over one billion model calls per day, is one of the largest language models actually writing the most production code in the world. We're also on our fourth or fifth generation of it. And it's trained using product data, of seeing where AI is helping people and where it isn't, trying to predict how it can help humans. It also requires a ton of infrastructure and specialty talent to be able to make those models really good. For instance, one of the people who has worked on those models with us is Jacob Jackson, who actually built GitHub Copilot before GitHub Copilot, which was called TabNine and was the first type of programming autocomplete product. He's also one of the people who built one of the first million token-context window models, and so he has done a lot of work on making models understand more and more and more information, and yeah, specialty talent and specialty infrastructure, too, to do that work. I think that in our ambling, kind of winding way to working on Cursor, one of the things that really did help us was when we were working on CAD and also in some of our explorations before, my cofounders had to dig very deep into the machine-learning infrastructure and modeling side of things. When we actually set out to work on Cursor, we thought it would be a long time before we started to do our own modeling as product lovers, but it happened much sooner than we expected. Recently, I had dinner with the CTO of a Big Tech company, and I asked him about what coding tools were popular with his engineers, and he told me that he regularly surveys them on this question, and they had Cursor available as a trial. He said he was getting these panic messages from engineers saying, 'Please tell us you're not about to take away Cursor,' because they'd become so dependent on it. Can you give us a sense of why, for programmers, this has kind of felt like a before-and- after moment in the history of the profession? What is it that tools like Cursor are making so different in the day-to-day lives of these engineers? I think that we're just already at a point where we are far, far, far from the ceiling of where things can go, and far, far, far from a world where much of coding has been replaced with something better. But just now at this point, these products and these models can do a lot for programmers and are already taking on quite a bit of work. I think the technology is especially good for programming for a few reasons. One is that programming is text-based and that is the modality that the field has figured out perhaps the a lot of programming data on the internet too, so a lot of open-source code. Programming is also pretty verifiable. And so, one of the important engines of AI progress has been training models to predict the next word on the internet and making those models bigger. That engine of progress has largely run its course; there's still more to do there. But the next thing that's kind of picked up the torch in making models better has been reinforcement learning. So it's been basically teaching models to play games, kind of similar to how in the mid-2010s we, humanity, figured out how to make computers really good at playing Go and Dota and other video games. We're kind of getting to a level of language models where they can do tasks, and you can set up games for them to get even better at those tasks. And programming is great for that, because you can write the code and then you can run it and see the output and decide if it's actually what you want. And so I think there's a lot about the technology that makes it especially good for programming, and, yeah, it's just I think one of the use cases that's the furthest ahead in deploying this tech out to the world and people finding real value from it. My sense is, maybe if I used to have to work eight hours a day, now it's maybe closer to five or six. Is that part of it? I think yes, in the sense that I think that the productivity gains of what would have taken you eight hours before in some companies now actually can take you five or six hours. I think that that is real, not across all companies, but it is really real in some companies. But what I would nitpick on there is I don't think programmers are shortening the hours they're working. I think a lot of that is because there is just a ton of elasticity with software, and I think it's really easy for people who are nontechnical, or who just don't program professionally, to underrate how inefficient programming is at a professional scale, and a lot of that is because programming is kind of invisible. Consider what programmers are doing at a company like Salesforce, where there are just tens of millions of lines, many millions of files of existing logic that describe how its software works. Anytime they have to make a change to that, they have to take that ball of mud, that massive thing that is very unwieldy, and they need to edit it. That's why I think that it's just kind of shocking to many people that some software release cycles are so slow. So yes, I think that there are real productivity gains, but I think that it's probably not reducing the number of hours that programmers are working right now. All right. Well, you mentioned nontechnical people. Cursor is used by a lot of professional programmers, but this year saw the coining of the term 'vibe coding' to describe what more amateur programmers can do, sometimes even complete novices, and often with tools like Cursor. How big is the vibe-coding use case at Cursor and what do you think is the future of vibe coding? So our main goal is to help people who build software for a living, and for right now that means engineers, and so that's our main use case. It's been interesting to see as you focus on that use case and use the understanding you get from it to push the tech forward and hop up programmers to ever-higher levels of abstraction, how it then also makes things more accessible, and that's something that we're really excited about. I think in the end state, building software is going to be way more accessible. You're not going to have to have tons of experience in understanding programming languages and compilers. But I do think that we're a decent bit away from a world where anyone can do this. I think there's still a bunch more work to do before anyone can build professional-grade software. That said, it's been really cool seeing people spin up projects and prototypes from scratch, seeing designers in professional settings doing that. It's been really interesting to see nontechnical people contribute small patches and bug fixes or small feature changes to professional software projects already. And that's kind of the vibe-coding use case, not our main use case, not where the company makes most of its money, but one that I think will become bigger and bigger as you push the ceiling of focusing on professional developers. I'm curious what you think of as the demand for it, though. I understand it's not your focus of the business. People like to talk about it, and I think it feels cool to have never built software before, and all of a sudden the next thing you know, you actually created a little to-do list app for yourself or something. Yes. I probably differ from some of my colleagues on this, where I think that, in the world as it exists right now, of the two buckets of that vibe-coding use case, there's an entertainment bucket if you're doing these things mostly for personal enjoyment or hobbies, and then there's a bucket that's more professional, and I think that that's designers doing prototypes or that's people who work to serve customers and are contributing back bug fixes to a professional code base. The way in which I probably differ from some of the people I work with is there's a group of people who are really, really, really interested in end-user programming and throwaway apps and personalized software, where everyone entirely builds their own tools. And I think that that's really cool. I think enabling that is really cool, and I think some people, a lot of people who aren't technical will be interested in doing that. But I still think even if you get to a world where anyone can build things on computers, I think most of the use cases will still be served by a small minority of 5 percent of the world that cares a ton about the tools and building them, and that everyone will use those tools more, because I just think that the interest in that stuff really differs among the population. So yeah, right now commercially I think that a lot of the more vibe-coding stuff falls more into a midjourney camp or an entertainment camp. It's something that some people get interested in for a bit and then kind of put it aside. And then some of it is in this professional camp of people who work on software for a living but don't code right now. I think you're right, because when I worked at more traditional companies, whenever a new piece of software was introduced, everyone would get upset. So that's my case for most people not becoming pro-vibe coders. I like software though, so I'm vibe-code curious. Maybe two or three generations from now in Cursor I'll be able to make myself something useful. You mentioned earlier that there are these two main ways that people use Cursor. There is the 'I'm looking at code and you're helping me autocomplete things,' and then there is the 'I'm going to give you a task and walk away and come back and see what you've built.' You told Stratechery's Ben Thompson recently that over the course of the next six to 12 months, you think you can get to a place where maybe 20 or 25 percent of a professional software engineer's job might be the latter use case of just handing off work to the computer and having the computer do the work end to end. Do you have any updates to that number in the past month or so? How high do you think that number can scale, ultimately? I think these things are really hard to predict. Yeah, I think there are some things that are blocking you from getting to 100 percent. One is having the models learn new things, like understanding an entire code base, understanding the context of an organization while learning from the mistakes. And I still think that the field doesn't have an amazing solution for that. There are two candidate solutions. One is you make the 'context windows' longer, which is that these large language models have a fixed window of text or images that they can see, and then there's a limit to that. Outside of that, it's just the model that came off the assembly line and then that new kind of information that's put into the model's head, which is very different from that of humans because humans are going through the world and your brain is changing all the time, you're getting new things that kind of persist with you, and obviously some memories fade away but persist with you somewhat. So candidate solution number one to the continual learning problem is just make the context windows really big. Candidate solution number two is to train the models. So every time you want them to learn a new thing or a new capability, you go and collect some training data on that, and then you throw it into the model's mix. Both of those have big issues, I think, but that's one thing that's stopping you. I think that the rate of really consequential ideas in machine learning that are new paradigm shifts is pretty low industrywide, even though the rate of progress has been really fast over the past five years. So, ideas in the form of replacing long context or in-context learning and fine-tuning with some other way of continual learning, I don't think that the field actually has an amazing track record of generating lots of ideas like that. I think ideas like that come about at the rate of maybe one every three years. So I think that will take some time. I think the multimodal stuff will take time too. The reason that's important for programming is you want to play with the software, and you want to be able to click buttons and actually use the output. You want to be able to use tools also to help you make software, tools that have GUIs. So, for instance, observability solutions, like Datadog, are important for understanding how to improve a professional piece of software, so that feels like it's needed. These models can also work coherently for minutes at a time, now even hours in some cases, but it's a different thing to work on a task for the equivalent of weeks in human time. So, just even architecturally, knowing if we're going to be coherent over sequences that long will be interesting to see, and that I think will be tricky. But there are all of these technical blockers to getting to something that's 100 percent, and there's many more that you could list and there are also many unknown unknowns. I think that in a year or so, even with just going from a high-level text instruction to changes throughout a code base, I think in the bull case you could probably do over half of programming as it exists today. I see these studies that Meter puts out where they look at the average length of time that a software or an AI model can do, and it does keep doubling at this really impressive rate. So, I think the hurdles that you identify are super important, but when you pull it back, it does seem like the task is really improving. Ultimately, humans don't tend to work on discrete tasks that are all that long. So I do think it's getting easier for people to imagine a full day's work. Definitely, definitely. I think that just forecasting these things is tricky, but one related field that can maybe foretell how things will evolve here is the history of self-driving, which has obviously advanced in leaps and bounds. In San Francisco, there are Waymos, which are commercial self-driving cars, and my understanding is that Tesla has also made big improvements. But I remember back in 2017, when people thought self-driving was going to be done and deployed within a year. Obviously, there are still big barriers to getting it out into the world. As hard and varied as self-driving is, it does seem like a much lower-ceiling task than some of the other stuff that people in the field are talking about right now. So we will see. I do want to ask you about the timeline, but I'm going to wait until a little bit later. All right, let me now ask you some of the famous Decoder questions, Michael. How big is Anysphere today? How many employees do you have? We're roughly 150 people right now. Okay, and when you think about how big you want the company to be, are you somebody who envisions a very big workforce? Or do you see a smaller, nimbler team? We do like a nimbler team, and I think the caveat here is while we want to keep the team nimbler for the scope of work that we're tackling, it will still mean growing the team a lot over the next couple of years. But yeah, I wonder if it will be possible to build a thriving technology company that does really important work with a maximum team size of maybe 2,000 people, or something like that. Something of the size of The New York Times. We're excited to see if that is possible, but we definitely need to grow a lot more from our current head count. What is your organization chart like? You have a few cofounders. How do you all divvy up your responsibilities? The two biggest areas of the org are engineering and the research side of things, like R&D generally, and then the go-to-market side of things, like serving customers. And this is a company that has really benefited from having a big set of cofounders and a big, very capable founding team. And so there's a lot of dividing and conquering across that scope. In particular, we've had an important group of people on the founding team who've done phenomenal work in building out that early go-to-market side of things. A lot of that comes entirely from the founding team, and is entirely credited to a subset of it. And so there's a lot of dividing and conquering across the business. At the same time, I think once you zoom in to the technical side of things, there's an intense focus from the four cofounders on that, and putting all the eggs in that one basket. I think we're lucky enough to be at a time when there are really, really useful products to build in our space. And I believe that the highest order of it, the thing you cannot mess up, is producing the best product in the space. And so we've been able to stay relatively lean in other parts of the business, especially relative to our scale, but also as a ratio to engineering and research, and still be able to grow. What part of the business do you keep for yourself? Where are you getting your hands dirty, and where would you get mad if someone tried to take that away from you? I spend a lot of time doing what I can to help grow the team. We think hiring is incredibly important, especially the hiring of ICs [individual contributors]. I think that one way technology companies die is that the best ICs start to feel disengaged, that they don't have control over the company, and talent density lowers. If you're working on technology, no matter how good the management layer is, if you have less than excellent people doing the real work, I think there's only so much you can do. I think that the dynamic range of what management can do becomes kind of limited. So l help by devoting a bunch of time to hiring. We actually got to maybe 75 people with just the cofounders hiring without engaging functional recruiters. Now I have fantastic people helping us with hiring. I have people on the recruiting side who work with us closely. But I spend a bunch of time on that and then try to help however I can on the engineering and product are the two biggest areas of focus, and then there's a long list of long-tail things. So you're fairly young, I think you're 25, and have had to make a lot of really big decisions about raising money, making acquisitions, all those hiring decisions that you just made. How do you make decisions? Do you have a framework that you use or is everything ad hoc? I'm not sure there's one framework. Some pretty common strategies that help us are, we try our best to farm all up and down the group, the org. This is not just for me — we try to do this for all decisions in the company. We increasingly have a very clear DRI [directly responsible individual], and then lots of other people offer their input. Every decision is pretty unique. Other devices that are well-known and have helped include understanding how high stakes and reversible the decision is. And I think that especially when you're in a vertical like ours, given the speed that it's moving, there's just a limit on the amount of time and the amount of information you can gather on each thing. Yeah, and then other devices, like clearly communicating the decision and using that as a way to force clarity for how it was thought through. Well, let's talk a little bit more about hiring, since you brought it up. There has been talk that OpenAI had considered acquiring you. I have to ask, given his recent spending spree, has Mark Zuckerberg invited you to his house in Tahoe? [Laughs] No, no. No? He's not coming around with his $200 million signing bonuses saying, 'Michael, why don't you kind of come over here? We're building super intelligence?' No. This for us is kind of life's work territory. So yeah, we feel really lucky to have the technology lineup, the initial founding team lineup, the people who have decided to join us, the way things have gone on the product to have the pieces in place to execute on this ambitious goal of automating programming. And time will tell if we're going to be the ones to do that, but as people who have been programming for a long time and working on AI for almost as long, being able to reinvent programming and help people build whatever they want to on computers with AI, kind of feels perfect for us. It feels like one of the best commercial applications of this technology too. So I think that if you can succeed in that, you can also push the field forward in big ways for other verticals and other industries. And so, no. Yeah, it sounds like you really want to stay independent. Yeah. Has Meta's recent hiring spree made it noticeably harder for you to recruit lately? No, not really. We try to keep the research team fairly small. I mean, the whole company is kind of small relative to what it's doing, but especially the research team. I think that people think through hiring decisions in different ways, and I think what we have to offer is most appealing to people who want to be a part of an especially small team working on something focused, that's solving problems with AI out in the real world. We're kind of this weird company. You talked about some products that are being made by some of the great folks who work on the API models. But I think we're this weird experiment of a company that's smack dab in between the foundation model labs and normal software companies; we try to be really excellent at both the product side of things and the model side of things and have those feed into each other. And so we appeal to I think a certain type of machine-learning researcher or ML engineer. And for them, I think it's about being part of this, and a little bit less about being part of some of the other things. One last hiring question. It was reported this week that two folks who used to run Claude Code whom you'd recruited to come over to Cursor left after a couple of weeks. Can you speak at all to what happened there? Cat [Wu] and Boris [Cherny] are awesome, and I think that they have a lot left to do on Claude Code, and they're really, as I understand it, the people behind that and that is their creation. As someone who's been working on something for three and a half years since inception, I understand the ownership that comes with that. I think that they have a lot left to do and they were excited about that, and so they've decided to stay [at Anthropic]. It seems that you were mentioning this interesting position Cursor sits in, in between the big labs and other startup companies that are using your software. How do you describe Cursor's culture when you're recruiting people? I think that some of the things that describe the current group, perhaps unsurprisingly — we are process skeptical and hierarchy skeptical. So, as we take on more and more ambitious projects, more and more coordination is required. But at a certain level, given the scope of the company, we try to stay pretty light on each of those. I think it's a very intellectually honest group, where people feel comfortable. It feels very low stakes to criticize things and just be open when giving feedback on work. But I also think it's a very intellectually curious group. I think people are interested in doing this work for the end goal of automating programming — separate from any work-life balance issues, because we want this to be a place where people at all levels of work-life balance can do great work. It's a place where so far no one really treats it as just a job. They're really, really excited about what we're doing, and I think it's kind of a special time to be building technology. I think to the outside world, what we do seems very focused and understated, partially because of how little communication we have with the outside world. We need to get much better at that. I think for the most part people think of Cursor as, 'Oh, that thing that grew really fast.' They know about top-level metrics and things like that to gauge just how fast the adoption has been. Internally, we've thought that it's really important to hire people who, while they might be very ambitious, are still very humble and understated and focused and level-headed, because there's noise left and right. I think that just having a clear focus and putting your head down are actually really, really important not only for people to be happy in this space but also for the team's execution. You mentioned communicating with the outside world. I think Cursor's history is mostly just a history of delighting its customers. But you did have this moment recently where you changed the way you price things, and folks got pretty mad. Basically, you moved from a set fee to more usage-based pricing, and some people ran over their limits without realizing it. What did you learn from that experience? I think that there was a lot to learn from that, and a lot on our end that we need to improve. To set the stage, the way Cursor pricing has worked, even back when Cursor first started, is by and large, you sign up for a subscription, and then you get an allotment of a certain number of times you can use the AI over the course of your subscription term. And the pricing has evolved. Features have been added, features have been changed, kind of up and down that limit, and there have been different ways you could pay down that limit or not pay down that limit over time. What's happened in parallel is using the AI once, and what that means is the value that gives people and the underlying costs in some cases have changed a lot. One big switch there for us is that increasingly when 'you use the AI,' the AI's working for longer and longer and longer. So you called out that chart that you've seen where it shows the max time that AI can work, and it's gone from seconds to minutes to hours at this point, and it's gone up very fast. We're on the front lines of that, where now when you ask the AI to go do something or answer a question, it can work for a very, very, very long time. That changes the value it can give to you. You can go from just asking a simple programming question to having it write 300 lines of code for you, and that also changes the underlying costs. In particular, it changes less the median and more the variance of those costs. So we bundled together a series of pricing changes, and the one that garnered the most attention was switching from a world where the monthly allotment is in requests to one where it's in the underlying compute that you're spending. One thing to knit on what you said is that usage-based pricing had been a big component of Cursor before, because over the life of Cursor, people have used the AI more and more and more and more. And then they started running out of limits, and we wanted to give people a way to burst past that. What this did is it changed the structure of how that usage pricing worked, where it's not on a request basis but on the underlying compute basis. That definitely could have been communicated legions better. I think that there's a lot we learned from that experience, and a lot we need to improve on in the future. I think it's hard for consumers in particular to understand usage-based pricing, because they're used to Spotify and Netflix, where they pay their 10 or 20 bucks a month and it's sort of all you can eat. The economics of AI don't really work that way. Yeah, I think that it will be interesting to see how things play out in our space in particular, because I think that for the consumer chat-app market, so far at least, it would be interesting to see how the curves of just how compute per user over time have gone up. But I wouldn't be that surprised if it's been pretty flat over the past 18 months or so, where the original GPT-4, I'm not privy to any inside information, but it seems like there have been big gains from a model-size perspective, where you can actually miniaturize models and get the same level of intelligence. And so I think that the model that most professional users are using in something like a ChatGPT has actually maybe gotten smaller over time; compute usage has gone down. But in our space, I think that for one user, compute is probably going to go up. There's a world in which the token costs don't go down fast enough, and it starts to become a little bit more like AWS costs and a little bit less like Percy productivity software, and that still remains to be seen. But one thing to note is that we do think it's really, really, really important to offer users choice, and so we want to be the best way to code with AI, if you want to turn on all the dials and get the best, most expensive experience. We also want to be the best way to code with AI if you want to just pay for a predictable subscription and get the best of what that price can offer you. And even for the main individual plan, the $20 Pro plan, the vast majority of those users don't hit their monthly limits, and so aren't hit with a message saying they need to turn on usage pricing, or not. That's the kind of AI user I am. I never hit it, which makes me feel that I need to be using it more. There is a really, really big difference between the top 5 percent and a median user. So some people are very, very, very AI forward. Well, coming into my last couple of questions here, I want to try to get at how AGI-pilled you are, because when we were talking earlier, you're sort of identifying all these very real technical problems in building more advanced systems that aren't just truly unsolved problems in AI. The size of the context when giving these systems longer memory, helping them learn the way that a human might be able to learn, we don't know how to do that yet. Yet there are lots of folks in the industry who believe that by 2027, 2028, the world will look very, very different. So, where do you sort of plot yourself on the spectrum of people who believe that everything is absolutely about to change, and we're sort of at the start of a process that's going to take decades? I think we're kind of on this bet in the messy middle, where we do think it's going to take decades. We do think that nonetheless, AI is going to be this transformational technological shift for the world. Bigger than maybe... just a very, very, very big technological shift. And when we started working on Cursor, it was funny, we would get these dual responses, and I think one is now increasingly falling out of favor with the rise of the first AI products that have reached billions of people. But early in 2022, we would get two reactions. One reaction was, 'Why are you working on AI? I'm not sure that there's really much to do there.' And then the other reaction we'd get, because we did have close friends and colleagues who were very interested in AI, was, 'Why are you working on 'insert X' application' — whether it be CAD or whether it be programming specifically — 'when AGI is going to wipe all of this stuff out in Y years,' maybe in 2024 or 2025. We think it's this middle road of this jagged peak, where if you actually peek under the hood at what's driven AI progress so far, I think that, again, there's been a few ideas that have really worked, there's been lots of details to fill in between, but there have been a few really, really important ideas. I think that despite the number of people who have worked on deep learning over the past decade and a half, the rate of idea generation in the field — really, really consequential idea generation in the field — hasn't budged that much. I think that there are lots of real technical problems that we need to grapple with. So, I think that there's this urge to anthropomorphize these models and see them be amazing and human or even superhuman at some things, and then think that they will just kind of be great at everything. I really think it's this very jagged peak. So, I think it's going to take decades. I think it's going to be progressive. I think that one of our most ambitious hopes with Cursor is if we are to succeed in automating programming and building an amazing product that makes it so you can build things on computers with just the minimal intent necessary, maybe the success of that and the techniques that we need to figure out in doing that can also be helpful for pushing AI forward and pushing progress forward in general. I think the experiment to play back here is if you were in the year 2000 or 1999 and you wanted to push forward with AI, one of the best things you could do is work on something that looks like Google, and make that successful and make that R&D available to the world. So, in some ways at least, I think about what we're doing is trying to do just that. So it sounds like you don't think that there's just going to be one big new training run with a lot more parameters and we're going to wake up to a machine god. Time will tell. I think it's important to have healthy skepticism about how much you can know with these things. But my best guess is that it will take longer than that, yet also still be this big transformational thing. All right, well, last question here. We've talked a couple of times today about how hard predictions are in general, so I'm not going to ask you to do something crazy like predict what Cursor is going to look like five years from now. But when you think about it maybe two years from now, what do you hope it's doing that it isn't quite doing yet? I think a bunch of things. So I think in the short term, we're excited about a world where you can delegate more and more work to very fast, helpful humans, and you can build a really amazing experience for making that work delightful while orchestrating work among these agents. Another idea that we've been interested in for a long time, which is a bit risky, is if you can get to a world where you're delegating more and more work to the AI, you'll start to run into an issue, which is whether you look at the code. And are you reading everything line by line, or are you just kind of ignoring the code wholesale? I think that neither closing your eyes and ignoring the code entirely in a professional setting nor reading everything line by line will really work. So, I think you'll need this middle ground, and I think that that could look like the evolution of programming languages to become higher level and less formal. All that a programming language really is is a UI for you as a programmer to specify exactly what you want the computer to do. And it's also a way for you to look at and read exactly how the software works right now. I think that there's a world where programming languages will evolve to be much higher level and more compressed. Instead of millions of lines, it's hundreds of thousands of lines of code. I think that for a while, an important way you build software is you could read, point at, and edit that kind of higher-level programming language. That also gets at a bigger idea that's behind the company: there's all this work to do on the model side of things. The field's going to do some of that, and we're going to try to do some of that. But then the end state of what we want to do is also this UI problem of how we get the stuff that's in your head onto the screen. I think that the vision of you entirely building software by just typing into a chat box is powerful. I think that that's a really simple UI. You can get very far with that, but I don't think it can be the end state. You need more control when you're building professional software. And so you need to be able to point at different elements on the screen and be able to dive into the tiniest detail and change a few pixels. You also need to be able to point at parts of the logic and understand exactly how the software works and be able to edit something very, very fine-grained. That requires rethinking new UIs for these things, and the UI for that right now is programming languages. So I think that they're going to evolve. All right. Well, a lot of fascinating things that you're working on. Michael, thank you for coming on Decoder. Thank you for having me. Questions or comments about this episode? Hit us up at decoder@ We really do read every email! A podcast from The Verge about big ideas and other problems. Posts from this author will be added to your daily email digest and your homepage feed. See All by Casey Newton Posts from this topic will be added to your daily email digest and your homepage feed. See All AI Posts from this topic will be added to your daily email digest and your homepage feed. See All Business Posts from this topic will be added to your daily email digest and your homepage feed. See All Decoder Posts from this topic will be added to your daily email digest and your homepage feed. See All Podcasts Posts from this topic will be added to your daily email digest and your homepage feed. See All Tech
New York Times
25-07-2025
- Politics
- New York Times
Trump Fights ‘Woke' A.I. + We Hear Out Our Critics
Hosted by Kevin Roose and Casey Newton Produced by Rachel Cohn and Whitney Jones Edited by Jen Poyant Engineered by Katie McMurran Original music by Dan PowellElisheba IttoopMarion Lozano and Rowan Niemisto On Wednesday, President Trump signed three A.I.-related executive orders, and the White House released 'America's A.I. Action Plan.' We break down what's in them, how the federal government intends to target 'political bias' in chatbot output, and whether anyone will stand up against it. Then, do we hype up A.I. too much? Are we downplaying potential harms? We reached out to several prominent researchers and writers and asked for their critiques about how we cover A.I. For a limited time, you can get a special-edition 'Hard Fork' hat when you purchase an annual New York Times Audio subscription for the first time. Get your hat at Guests: Brian Merchant, author of the book and newsletter 'Blood in the Machine' Alison Gopnik, professor at the University of California, Berkeley Ross Douthat, New York Times opinion columnist and host of the podcast 'Interesting Times' Claire Leibowicz, head of A.I. and media integrity at the Partnership on AI Max Read, author of the newsletter 'Read Max' Additional Reading: Trump Plans to Give A.I. Developers a Free Hand The Chatbot Culture Wars Are Here 'Hard Fork' is hosted by Kevin Roose and Casey Newton and produced by Whitney Jones and Rachel Cohn. We're edited by Jen Poyant. Engineering by Katie McMurran and original music by Dan Powell, Elisheba Ittoop, Marion Lozano and Rowan Niemisto. Fact-checking by Caitlin Love. Special thanks to Paula Szuchman, Pui-Wing Tam, Dahlia Haddad and Jeffrey Miranda.
Yahoo
23-07-2025
- Business
- Yahoo
Hard Fork
Credit - The New York Times' weekly technology podcast, hosted by journalists Kevin Roose and Casey Newton, covers topics that impact our daily lives without veering into wonky debates over the specifics of a new iPhone rollout. Topics include how AI is impacting the job market for new graduates looking for entry-level positions, the ethical hazards of an AI chatbot that's too nice, and the Trump phone. Episodes strike the right balance between interviews and commentary, and when the co-hosts score big execs on the podcast, they question them directly on subjects like the speed with which Silicon Valley seems to want to leap into the world of artificial intelligence without guardrails. The duo recently made headlines for their live interview with OpenAI CEO Sam Altman who sparred with the journalists over the Times' ongoing lawsuit against OpenAI. Write to Eliana Dockterman at
New York Times
18-07-2025
- Business
- New York Times
X Hits Grok Bottom + More A.I. Talent Wars + ‘Crypto Week'
Hosted by Kevin Roose and Casey Newton Produced by Whitney Jones and Rachel Cohn Edited by Jen Poyant Engineered by Katie McMurran Original music by Dan PowellMarion LozanoRowan Niemisto and Alyssa Moxley This week, we tick through the many dramatic headlines surrounding xAI, including the departure of X's chief executive, Linda Yaccarino; the Grok chatbot spewing antisemitic comments; and the A.I. companion Ani engaging in sexually explicit role-play. Then, we explain why a fight to acquire the start-up Windsurf startled many in Silicon Valley and may reshape the culture in many of the big A.I. labs. And finally, it's 'crypto week.' David Yaffe-Bellany explains how crypto provisions in the bills before Congress and the president could affect even people who don't hold digital currencies. Also, we officially have merch! For a limited time, you can get a special-edition 'Hard Fork' hat when you purchase an annual New York Times Audio subscription for the first time. Get your hat at Guests: David Yaffe-Bellany, New York Times technology reporter covering the crypto industry Additional Reading: Elon Musk's Grok Chatbot Shares Antisemitic Posts on X Google Hires A.I. Leaders From a Start-Up Courted by OpenAI Cognition AI Buys Windsurf as A.I. Frenzy Escalates 'Crypto Week' Is Back on Track After House G.O.P. Quells Conservative Revolt The 'Trump Pump': How Crypto Lobbying Won Over a President 'Hard Fork' is hosted by Kevin Roose and Casey Newton and produced by Whitney Jones and Rachel Cohn. We're edited by Jen Poyant. Engineering by Katie McMurran and original music by Dan Powell, Marion Lozano, Rowan Niemisto and Alyssa Moxley. Fact-checking by Caitlin Love. Special thanks to Paula Szuchman, Pui-Wing Tam, Dahlia Haddad and Jeffrey Miranda.
