Building an AI-Powered Terminal | Zach Lloyd

Zach Lloyd is the cofounder and CEO of Warp, the intelligent terminal powered by AI. They recently raised their $50M Series B led by Sequoia Capital. He was previously the cofounder of SelfMade and the interim CTO of TIME.

Zach's favorite book: The Death of Ivan Ilyich (Author: Leo Tolstoy)

(00:01) Introduction
(00:07) Basics of the Terminal: What It Does and Why Developers Use It
(01:48) Foundation of Warp: Addressing Terminal Shortcomings
(05:00) Initial Feature Set: Building Warp’s Early Product
(07:19) Product Iteration and Retaining Early Users
(10:15) AI Integration in Warp: Initial Use Cases and Evolution
(14:00) Technical Challenges: Building Warp in Rust
(16:25) Feature Prioritization: Balancing Feedback and Vision
(18:26) Warp's Tech Stack: Languages, Frameworks, and Tools
(22:22) Cross-Platform Development Challenges
(23:07) Importance of Design: Competitive Advantage in Dev Tools
(26:10) Hiring for Design in Early-Stage Startups
(30:22) Warp’s Vision for the Next Five Years
(32:15) Lessons from Scaling Warp: Advice to Younger Self
(34:02) Rapid Fire Round

--------
Where to find Zach Lloyd:

LinkedIn: https://www.linkedin.com/in/zachlloyd/

--------
Where to find Prateek Joshi:

Newsletter: https://prateekjoshi.substack.com 
Website: https://prateekj.com 
LinkedIn: https://www.linkedin.com/in/prateek-joshi-91047b19 
Twitter: https://twitter.com/prateekvjoshi 

Transcript

Prateek Joshi (00:01.432)
Zach, thank you so much for joining me today.

Zach (00:04.818)
Thanks for having me. I'm excited to be here.

Prateek Joshi (00:07.616)
Let's start with the basics. For people who may not know, can you explain what the terminal does?

Zach (00:16.986)
Sure. So the easiest way to explain it is if you've ever seen a movie where there's a hacker and they're trying to break into a system or diffuse a bomb or something like that, you'll see they pull up this application as like a black background and green text. And what you do in that application, unlike most other apps you use on your computer, is you type commands and the commands run.

computer programs. so, you know, most of the population of the world has moved on to use more graphical interfaces to do all their, you know, that's how they use computers, but developers in particular, still use the terminal in this text based interface for interacting with computers for lots and lots of their tasks. And they, you know, they do that because

A, it's like a very efficient interface, just like you type, something happens. It's a very flexible interface compared to like a GUI. Like if you're just dealing with text, you don't have to like define buttons and interactions and animations and that kind of stuff. So it's a really, really flexible interface. And developers in particular will do it for all sorts of things, whether it's writing their code or deploying the code to live on a server someplace.

or debugging, all sorts of tasks. So that's the terminal, very high level.

Prateek Joshi (01:48.322)
That's a great example. And I think every single person can relate to that because we've all watched those movies where, you know, the character brings up like an amazingly cool, like black screen, that type of bunch of stuff. And, you know, cool things happen. And this forms the basis, the foundation for Warp, the company you founded, what you're working on. So let's start with, you know, what are the shortcomings of this terminal?

Zach (02:02.393)
Yes.

Prateek Joshi (02:17.843)
user experience. Like what parts are broken and what led you to to launch Warp?

Zach (02:24.078)
Yep. So when you use a traditional terminal, let's say predating warp, the experience is a bit like stepping into a time machine. Like when you open this app, it's not like any other app on your computer. And so just to give you examples, like in other apps, you can use the mouse to do things after you type in text. So you can select the text, you can hit like...

you know, command X, can cut, copy, paste it, you can put the cursor someplace, you can start typing. The terminal doesn't let you do any of that, the normal terminal. What is actually happening is when you run a terminal, like again, before work, the program is essentially copying the behavior of a piece of hardware that hasn't been widely produced since the 80s. So you gotta kind of like use your imagination and like go back to a movie where it's like,

you know, it's the green screen thing. that hardware predates like the mouse working. And so there's really basic weird stuff like that. And then it's copying this really old piece of hardware. like I said, it's kind of like stepping back in time. The other sort of like problem with it is that because it's all text based, you know, there's a really steep learning curve and you really have to like...

remember a lot in order to do anything with it. So you need to remember these sort of like text based commands, which are not English words. They're like, you know, they're depending on what you're using. It's probably some variant of like a Unix command. So like you need to know how to like navigate your file system with them or search files or, you know, different developer tools have very specific commands and these commands can get really complicated in terms of the number of options that they have. And so

there's this demand put on the user to really learn this arcane language, which is hard. And I would say, especially for new developers, it's a really intimidating, not very welcoming first experience to becoming a programmer. So it's also very easy to mess something up. So if you type the wrong character or the wrong command, all of a sudden you've like,

Zach (04:46.81)
You've something, you've set some weird state in your terminal and good luck. You're off to stack overflows, right? And figure out how to undo it. Yeah.

Prateek Joshi (04:57.772)
Now, this command line, it's been around for decades. And obviously as engineers, we have used it like quite a lot. there's a certain user behavior that comes with it. So when you decided to build warp, how did you decide on the initial feature set? And also you had to get this into the hands of the initial users. So how did you decide on what goes into that first version of the product?

Zach (05:00.836)
Yes. Yes.

Zach (05:22.33)
Yep.

Zach (05:26.426)
Yeah. So out of the gate, we were primarily trying to fix the problems that we experience ourselves as developers. know, so I've been a developer for a really long time. I've always used the command line. I am not a command line expert. I am a person who uses it to the extent that I have to, get my job done. And so our first set of features were all around, can we make this feel more like a modern app? Can we improve the UX?

And we thought that there would be, you know, kind of like a wide, wide developer interests if we could produce something that was essentially a more usable, more usable version of it. And so for us, that was things like, can you build in auto completions to it? Like you can get auto completions in a terminal, but it doesn't come with them standard and in work. It does. Can you make the command entry experience?

feel like you're using a word processor or an IDE. So the mouse works and copy paste works. Can you make the output experience? And what I mean by that is when you run a command, it produces a whole bunch of texts. And in a normal terminal, the terminal has no idea what text is connected to what command. And warp are like, well, it can work more like a notebook where you run a command, you get an output, and the input and the output are connected. And that lets you do cool things.

as simple as like, I just want to copy the output of this one command, which is not a trivial thing to do in a normal terminal. So we really like, we were like, let's look at the basic UI primitives and try to rebuild them from the ground up just to make the typical workflows that developers do in the terminal a lot simpler. That's where we started.

Prateek Joshi (07:19.724)
As you continued to get your early users, some clearly have grown and you raised a big round. And in that process, can you talk about, talk through your approach of iterating on the product, what it was, what it is now, and how did you do it?

Zach (07:21.839)
Yeah.

Zach (07:26.617)
Yep.

Zach (07:34.138)
Yeah, yeah, for sure. Yeah. So, okay. So first we, the sort of first milestone was can we build something that we can use ourselves? Which again, this is a huge benefit to if you're a developer building a developer product, you get to use it. And so the first thing was like, can we make it our own daily driver? Even if we don't necessarily prefer it to the alternative tools, because it had like a bunch of bugs and issues and that type of stuff.

So that was step one. The second thing was, can we get anyone else to use it? And it wasn't about growing it at all, it was about like, can we get any other retained users? So the initial metric was like, can we get some people to retain on it? And the goal was really like, can we get a feedback loop going? Which I think is the most important thing early on. If you want the product to improve, you need people telling you.

you know, what their experience is like with it. And so we actually first tried this, like, that's the question is like, how do you find the people who want to try it? And so we, we tried kind of like with our network. When I say we, was like me and a few other engineers who were working on this initially with me. And like our network actually wasn't awesome. And so what we ended up doing to really get the feedback loop going was we, we posted a wait list for

for warp describing like with like a landing page describing what it was, how it was built, what the product was going to be like. We built that it just had an email like sign up form. We posted it on Hacker News, which again, hit or miss. think it's very, who knows if you're going to get traction on it. But we got to the top of Hacker News with a waitlist, which was very, very cool. And then in like the first day, we got 10,000 developers to sign up to try to sign it, which was to try and

like, you know, get early access, which was, you know, really good signal that there was demand for at least something like what we were doing. And then we sort of gradually let people off that waitlist and cohorts seeing what the reaction was, would they retain. And it was only once we got to a place where we were retaining users, and initially not a huge percent of users, because, you know, products are rough in the beginning.

Zach (10:00.75)
But once we did that we were retaining users, the mindset shifted a little bit more to like, how do we get more people in? How do we grow it? But yeah, the order was like us, initial feedback, retain users, and then try to figure out growth.

Prateek Joshi (10:15.402)
And when you look at all the things a terminal has to do, there are some things that are simple, meaning, hey, I want to see the list of files in this folder. The one command isn't much to do. And there are certain portions where AI can really make it faster, better, nicer. So when you look at all the little pieces of work that can be done by AI, what

Zach (10:20.943)
Yep.

Prateek Joshi (10:38.954)
was the lowest hanging fruit. Like how did you decide, okay, these are the things that AI can, you we can make it better.

Zach (10:45.85)
Yeah, so I can tell you how we started with AI. So AI came into warp about a year and a half after we started building it. We started warp 2020 and the huge advancements in LLMs, I want to say came like 2022, but we had something using Codex from OpenAI pretty early on. And the first use case, which we thought made the most sense was

translate natural language into command, which is still a very common thing that our users do. And so, for example, it's like, can in warp, you can say, start a Docker container using this Docker image on this port. it just will, even as you type it, it will generate the command for you. And that's pretty magic because that's saving you, that's saving the user either like,

the trip to Stack Overflow, pulling up the help content for Docker. it's actually LLMs are awesome at that. They're really good at that particular use case. So that's where we started. And then from there, went to, the next step for us was a chat experience, which I think a lot of companies did when Chat GPT came out. We like, well, we need to get chat into warp and like,

what we tried to do was make it really easy to attach different parts of your terminal session as context into the chat. And so the canonical use case here would be something like you run a command, it produces an error. You're like, I don't know how to solve this error. And so you attach the error to a chat and then you get a solution and then you copy paste the solution back into the terminal. So that was our second iteration.

And then we're now on our third iteration, which I think is by far the coolest, which is the terminal input itself in warp accepts either English or command. And the conversation with the AI happens directly in the terminal session. And it's becoming more and more autonomous where you're not just asking about the last command, you're asking it to do a whole task for you, set up a whole new like node project for me.

Zach (13:09.962)
set up a new server for me on Gcloud and it can just, you know, step by step do these things as you semi-supervise it and like eventually not really even have to supervise it that much at all. So it's cool. It's a very natural interface because it's already where developers are talking to computers for them to just talk to LLMs to help them talk to the computer.

Prateek Joshi (13:32.608)
Now, in all these, you've been building Warp for a while and you have gone through iterations and I on the surface it looks like it's a great product, we love it, but I'm sure in the background there must have been a couple of big technical challenges you've had to overcome to make it look nice and cool and fun. can you talk about what is the biggest technical challenge you had to overcome to make it very nice and seamless for the user?

Zach (13:40.388)
Yes.

Zach (14:00.654)
Yeah, so we built Warp the hard way, I would say. So we experimented with an electron version of Warp. building Warp in JavaScript and like, if you build Warp in, you know, if you build an app in JavaScript or TypeScript or something like that, you get the benefit of this gigantic ecosystem and it can be cross-platform out of the gates and like get the browser rendering engine. But it was too slow.

Terminals are technically challenging because they need to have this sort of like huge throughput of text potentially. Developers are using this tool because they want speed. And so if it's laggy or if it's dropping frames or the text doesn't flow through it, it's a real turnoff to professional developers, which is who we're building for. So we then rebuilt the app and

in Rust and Rust is like a much more challenging systems language. We do everything from the ground up, so to speak, meaning it's a GPU accelerated terminal and we literally have graphics code that we have written for rendering characters, rectangles, images to the screen.

Prateek Joshi (15:20.12)
Mm-hmm.

Zach (15:25.874)
And we have a whole framework that works on top of that as far as like, do you program it? And so we, you know, we put in a lot of extra work to try to get to a product that really, really works well. So building it in Rust and doing the whole thing, nuts to bolts was a super, super duper high technical lift for us. But I think, you know, it comes across in the product.

Prateek Joshi (15:47.702)
Right. Yeah. I think, as you said, developers, it's a tough audience and every like millisecond of lag and they're very, I'm sure you've been building and selling to them. So I think it's, very challenging and when it works, it's super sticky. Meaning once they like something, they'll just stick with it for a long time. So great. Now going forward at this scale, obviously you've gone from zero to now, which is great. And going forward, how do you...

prioritize features for the product. Can you talk about how is it customer feedback? Is it internal product taste or a combo of both?

Zach (16:25.292)
It's a combo of a bunch of things. it's super important what customers and developers, users are telling us that they want for sure. It is a bet based on how the underlying AI technology is changing is really shaping our roadmap a lot right now. I think we're at a moment where no one quite knows what

software development is going to look like in like a year, two years, three years. It's like, you know, and there's competing visions for what that future is going to look like. Like, are people going to be using terminals? Are they going to be using code editors? Are they going to be using some new workbench? Is it going to be agents or Devin or something like that who are doing the work? And so a lot of what we're trying to figure out is how do we navigate like an aim?

appropriately far into that future, while still iteratively meeting developers where they're at right now and just giving them what they want. The amount of AI announcements and, for lack of a better word, is actually way ahead of the adoption. And so there's a risk of you could overshoot where the AI stuff lands.

So we think that like, you want to be kind of right at like the, you know, the border where you're taking developers from their existing workflows that they've done for the last 10, 20 years, like building code in the terminal and, you know, kind of bringing them into the future where the AI can help them. So that's kind of the philosophy, but yeah, to answer your original question, it's user feedback, but it's also vision. And it's also a lot of times it is our personal taste because we were like,

We live in the product, we use it every day.

Prateek Joshi (18:26.286)
I want to go back to your comment on Rust. taking a step deeper, obviously, to build something like this, many people, many founders, they are building developer tools. So as a starting point, can you talk about what languages, frameworks, tools are used today in the warp stack? Like what are you using internally?

Zach (18:30.542)
Yes.

Zach (18:51.374)
Yeah, so I mean, the warp terminal itself is built almost entirely in Rust. The only parts of it that are not built in Rust are pieces that are platform specific. so warp runs on Mac, it runs on Linux, it's in private beta on Windows, and it runs on web. And so for each of those different platforms, there's different platform interface points like

the windowing system or the graphics drivers. so sometimes like we have some objective C code, we have some JavaScript code, but the client app is almost entirely written in Rust. For the server, we use Go. If I had do it again, I'd probably use Rust on the server. There's a penalty in terms of having two different languages, but Go has some nice things too around just like support on the different clouds and great library support. So we use Go on the server.

And then on the AI side, there's a variety of LLM. So we, we allow users to choose to some extent what LLM is backing the AI. The current best one for coding and terminal stuff is, is Sonnet, Sonnet 3.5, the Anthropic one. When we initially launched, we were using more open AI, but we also use a mix of models for different types of tasks. So.

The code generation we do has a separate sort of model path. And so it's, it's, you know, there's a bunch of engineering that goes into how do you provide the best like AI access essentially.

Prateek Joshi (20:33.14)
And to build and maintain like separate apps for Mac and Linux and Windows, how do you think about finding people who know how to do the Mac native stuff? Obviously like programming and building a product in the cloud, that's good, that's the base, but it has to work equally well across these different systems.

Zach (20:39.32)
Yep.

Zach (20:50.191)
Yep.

Zach (20:54.168)
Yeah, this is a challenge. okay, so by using Rust, we're fortunate that we're sharing like, I don't know, 95, 98 % of the code across all these platforms. So that's a huge lever. Now, the things that need to feel native, like only Mac has like Mac menus, for instance, right? And so like, you know, we have to do engineering work around that. The biggest challenge for us has been

Prateek Joshi (21:06.733)
Mm-hmm.

Zach (21:23.834)
sort of Linux and Windows because we are a as a group, we all use Macs for development. And so what we had to do was actually just get people Linux machines and get people Windows machines. And then for Windows in particular, it's actually very cool. We're able to sort of work with the folks at Microsoft. We're doing this right now to try to make the app have like native look and feel. And so for instance, they are

advising us on what's the right sort of font to use for the app for Windows. What should the icon look like? Because we like ported over our Mac icon and they're like, no, that's not the right, like, like Windows icons don't have this box thing. So, so, and we don't really, we're doing our best to kind of get outside help on it. And then the other big thing is like, you know, the people who use the app on these platforms give us a ton of feedback on how to, how to make that the app really do well on the platform.

Prateek Joshi (22:02.35)
Right, right.

Prateek Joshi (22:22.958)
I think that's a great, great point to mention where you chose Rust and 90, 95 % of the lift is done because many times in the early days, for example, I don't know, iOS was Android. It's like maintaining two different companies almost because everything was broken. Nothing was portable. And this is really great. Now, I want to talk a little bit about product design. All the companies when they, there's certain,

Zach (22:38.652)
Such a pain.

Prateek Joshi (22:52.116)
indescribable coolness to the good products and it's almost it comes down to good design. how did you think about product design? Did you bring someone on board? How much does that person or the group plays a role in shipping today?

Zach (23:07.642)
This is a great question. the first person literally who joined me at Warp was a designer, a product designer who I'd worked with at Google. And I think that was a great choice. To me, actually, it's like a competitive advantage to have good design for developers. I don't want to overgeneralize, but I feel like a lot of developer software is kind of built by developers for developers without a ton of design input, whereas

Even most enterprise software now has kind of been consumerized as far as the quality of it goes. I would not say that's totally the case yet with developer software. so one of the things that we're trying to do is build consumer grade, highly designed, highly polished experiences in warp. And I do think it makes a difference. It makes a difference in terms of the user love that you get. It makes a difference.

When the interactions are like, you can tell that, you know, we've thought about how to like really remove friction. And if you do that in like, you know, a hundred places in the app, it actually has like real time impact for people too. So first person who came on to work with was a designer. Design is like a function within warp at the same level as product and engineering. We have a design leader at the company.

So I like to emphasize also, as we do product development, the main thing actually that I want to see is design. We have this saying, mocks over docs. I don't really want people writing stuff up. I just really want to see what we're going to build, either via Figma, mock, or design, or something like that. It's the fastest way to get everyone aligned on what we're doing.

Prateek Joshi (25:02.478)
This is such a great important point you mentioned because many times in many early stage products, design gets ignored. many times it's like the great design may not really build a company, but bad design kind of certainly would tank it because people who likes using poorly designed products, we don't talk about it, but everyone loves a well designed product.

Zach (25:23.566)
Nah.

Yeah. The other thing I would say is like for a product like warp where the product is a piece of software, I think it really matters. Whereas like say it's like Uber, like I wouldn't say Uber's product is Uber's app. Uber's product is getting you from like point A to point B as quickly as possible. And so maybe it's a different calculus, but for us, the thing that we're really trying to get people to love is this piece of software that they're going to have open all day on their computer.

And so, yeah, I feel like the product design has to be awesome.

Prateek Joshi (26:00.072)
And the word design, it's a big term. And in the early stage, there are many different shapes and forms in which designers come. So if you were guiding a young early stage founder and they want to bring someone like a product designer, what should they look for? What does that person do on day one? And also, what questions can you ask them during maybe the interview process or getting to know them process?

Zach (26:10.745)
Yes.

Zach (26:26.574)
Yeah, so the things that I'm looking for, if I'm going to work with a designer, are the ability for that designer or desire to work in public and to work with a very tight iteration loop. And I also am looking for like, does the person have like, you know, great craft, meaning like, are they able to produce things that are functional and beautiful? Like I would say that's the table stakes, but the thing that I would really look for, especially in an early stage startup context is like,

Does this designer want to like kind of like take your ideas and go off for like a week and come back with a design? Or is this a person who's like gonna in the extreme case get into a a Fig Jam with you and be like, let's kind of like iterate on this right now and let me get back to you and show you like the thing. Well, here's what I'm thinking in an hour or two. And so I think that cycle and I think Figma has really changed this because

Back in the day when I was working with designers who were using tools like Sketch, it was very much like a private thing and I'll share it when it's ready to be shared. And now the speed of iteration in design has gotten so much better. That's the main thing that I'm looking for. I'm also looking for a designer who can produce options, which is kind of a crazy thing that you don't always get. so responsiveness to feedback, willingness to iterate.

And also just like great visual sense and great like, know, user first perspective. That's what I care about.

Prateek Joshi (28:02.22)
Yeah, because the most common mistake or oversight I see is people say, design is mostly picking the right color or shapes. But I think the much bigger thing is shaping the product experience because of, yeah, exactly. So how do you test for that? how do you test their ability to shape this flow?

Zach (28:15.809)
Absolutely.

Zach (28:22.522)
Oh, okay. It's a great question. So the actual process that we do, I don't know how in the weeds to get here, but it's like, so when we, when we interview a designer, there, there is like a bunch of preamble stuff where it's like presenting work that they've done. But the thing that actually matters at the end of the day is that we have them come work with us for a day. And so they get like a real part of the app, a feature that probably not doing right now, we pay them for it.

think that matters, like, so that it's like, they're doing it for real. And it's like, we're not like getting free labor out of it. So we pay them for it. We do a kickoff and then throughout the day, we iterate with them in Slack and see if they can actually work with us to get to the, to get to the well-designed thing. And it, you're right. It has very little to do with like the, how the graphics look. has, especially for a product like warp, it's like,

It's much more around thinking about like, what's the user trying to accomplish? What's like the, you know, the least like how's, how's the user going to discover that they can accomplish it actually is really important. What's like the least friction flow for them to accomplish that task. And so the designer really needs to be able to put themselves in the shoes of our user to be able to produce something, something valuable. Yeah.

Prateek Joshi (29:47.778)
This is actually really good because when I noticed little things like the well-designed products, they care about, hey, what is the piece of software doing when the user is waiting half a second before I show the output? Even like those little things, when every little thing shapes the experience of the user and good products do it in a certain way, bad products are, they don't think about those little things. I think it's very, very interesting. Going forward, as you look at Warp,

What will warp be in five years?

Zach (30:22.116)
Yes, it's a great question. look, our ambition is to be, you know, a tool or the tool that when developers open up their computer and start building software, whether it's shipping a feature, fixing a bug, setting up something in infrastructure, that they're able to do more and more of that work from within Warp. What that future looks like,

My best guess right now is that it's gonna be less of a developer opening a file and typing in a bunch of code and then opening a terminal and building and testing and running it, and more of like, you ask, I'm little tired of the word agent, but let's just say agent, you ask a smart assistant to get you going and you collaborate with that assistant on the work.

And who knows, maybe the models get like good enough that a lot of the drudgery, like the bad tasks that I don't want to do as a developer can be totally delegated to some set of agents. But the the tasks that I enjoy as a developer, which are like, you know, building awesome things for users, solving really hard puzzles are things that I get to do. And these, you know, the intelligence of the system really helps me do it.

I think warp will be not really a terminal. don't, again, I don't know that that distinction is going to exist, but it's, it's going to be the place where developers start these tasks of, of changing code or changing their systems. and so, you know, our vision is, is much more like, you know, how can we help a developer in a few years sit down and just like do their job.

Prateek Joshi (32:15.182)
Amazing. I love that vision. And one final question before we go to the rapid fire round. If you were to give advice to young Zach five years ago on day one of warp about the lessons from launching and scaling warp, what would that be?

Zach (32:36.036)
Great question.

So I would push myself to move faster, take more risks and make more mistakes. And this has been a big focus for us lately and so this is why I'm saying it now. But it's a lot of times it's better to put something out and have it not work than it is to keep it kind of close to your chest, work on it for a long time and then put it out. And so there've been a few things in the lifetime of our product where it's like,

I wish we just put something out and got more signal on it before we invested more resources so that we could go faster. And specifically like right now, if you're starting something, I just think like in the developer space particularly, the speed of change is so high and this week there's been crazy stuff launched from like OpenAI and like...

It's just like the rate of change is so high, you need to prioritize speed almost over everything else. But you also, if the thing you're building isn't useful or doesn't build the value, or doesn't position you to build like a long-term business, like you can't, that also obviously matters. So it's really hard, but yeah, I would give myself advice to take a little bit more risk, move a little faster, make some more mistakes.

Prateek Joshi (34:02.478)
Alright, with that we are at the rapid fire round. I'll ask a series of questions and would love to hear your answers in 15 seconds or less. You ready? Alright, question number one. What's your favorite book?

Zach (34:10.306)
Okay, yeah, let's do it.

Zach (34:15.738)
I'm gonna say The Death of Ivan Ilyich by Tolstoy. Is that amazing?

Prateek Joshi (34:19.924)
Yeah, amazing. I love Tolstoy and that's a separate conversation, but amazing choice. All right, next question. What has been an important but overlooked technology trend in the last 12 months?

Zach (34:34.97)
I'm not seeing people talk enough about voice control, especially with the importance of like prompting, having voice input and voice control feel like not talked enough about.

Prateek Joshi (34:49.55)
What company do you admire the most and why?

Zach (34:54.01)
I'm gonna say Apple. like, that's what I aspire to on the product side, just like the crafting quality of the products. I use so many of in my daily life. I've had to use non-Apple stuff and it kind of drives me crazy. So I'll go with Apple.

Prateek Joshi (35:05.838)
What's the one thing about dev tools that most people don't get?

Zach (35:12.986)
I think until recently, really hard to make money off DevTools. The advent or rise of AI in DevTools is like the best thing that has ever happened to it because there's just like such an obvious monetization play for them.

Prateek Joshi (35:28.342)
What separates great products from the merely good ones?

Zach (35:33.466)
I'll say in an AI world, I think that it's how little work can you make the user do? Like, can you just like almost do the whole thing for them and read their minds is what makes it great.

Prateek Joshi (35:45.334)
What have you changed your mind on recently?

Zach (35:48.506)
The power of agents, which just the speed of the model is what I'm seeing when I play with O-1. I was like a little skeptical, but now I'm...

Prateek Joshi (35:59.094)
What's your wildest prediction for the next 12 months?

Zach (36:02.808)
In the developer space, think it's that a majority of pull requests are going to be started by AI. Maybe not finished by it, but started.

Prateek Joshi (36:10.99)
Right, amazing. All right, final question. What's your number on advice to founders starting out today?

Zach (36:20.186)
It is to try to build something really big. Don't think too small. Like paradoxically doing niche ideas are actually harder to fund. They're not as fun. You can't work on them as long, harder to recruit people. So I would try and do something really big and world changing.

Prateek Joshi (36:38.382)
I love that advice. think it's very, it's not understood well, but paradoxically doing the more ambitious thing is easier because of all the things you said. can attract people and capital and you'll be thrilled and fired up to pursue that big vision. So.

Zach (36:53.626)
I think so. It's it's so much more motivating. Like it's so tempting to start with, I'm going to do like, I'm going to get two enterprise contracts for some SaaS thing. I, you probably can get traction faster that way, but it's not, it's not why.

Prateek Joshi (36:56.491)
Right, right.

Prateek Joshi (37:06.082)
Yeah. right. Zach, this has been an amazing episode. Loved your hard-earned insights. And obviously, Warp is a beautiful product growing fast. So thank you so much for coming onto the show and sharing your insights.

Zach (37:19.64)
Amazing, it was awesome to be here. Thank you for having me.