Console

Developer interviews

Interview with Zach Lloyd

Founder, Warp - GPU accelerated terminal.

What is Warp? Why did you build it?

Warp is a reinvention of the terminal.

I was really interested in building something that could potentially impact all developers. The terminal is one of the two tools where if you walk by a developer’s desk, you’re likely to see it open alongside a code editor. For some developers, the terminal still is their code editor. It’s a ubiquitous tool, but from a product perspective there are a number of issues with it.

Firstly, it is a tool that’s hard to learn and hard to configure. It’s hard to get really good at it. On the flip side, if you do manage to get really good at it, you unlock a lot of productivity. I’ve worked with engineers throughout my career who were good at it, and who have been able to do things that I just couldn’t do as an average user.

One of our product ideas was can you make that power accessible to all developers?

The second idea on the product side really stemmed from my time being as a Principal engineer on Google Docs. I used to be the tech lead for that project and helped build a lot of Google Sheets in particular. That experience made me realize how if you take an app that has traditionally been desktop, single-user, non-collaborative software, and make it cloud-native and collaborative, that always unlocks significant productivity gains.

The idea was to do something like that for the terminal, which is one of the last apps I can think of that’s both very widely used, but remains non-collaborative and not oriented towards teams.

Warp needs to be a very fast app, so we built it in Rust. It’s GPU accelerated. Today, it’s Mac only and is in public beta, so anyone can go download it and start using it.

We plan on supporting more platforms by taking that same Rust codebase and using it for Linux, Windows, and also for the web where we would do WebAssembly and WebGL rendering of Warp.

How is it different from any other terminal?

As a developer, the features you’re going to notice first are the way that input and output work.

In Warp, output is more like a data notebook where we connect your command with the output that it produced, and we let you navigate your terminal on a command by command basis. We call those command outputs ‘blocks’ because they logically separate commands; this lets you easily copy and paste something you did in the terminal, or make a permalink, so you can share an output with your team, or search within an output, or rerun an output.

In Warp we’ve taken over the way that input works and made it work much more like a normal code editor. For instance, there are basic things in the terminal that are very jarring to people who are new to it and are annoying to people who use it a lot. The terminal is not mouse accessible. You can’t click and put your cursor someplace, you can’t double click, select, and then delete text. You can’t do advanced things like multiple cursors.

We’ve introduced a text area that works like the text editor in VS Code. You can position the cursor, and all your normal keyboard shortcuts work. There’s no learning curve for executing commands.

The third thing that’s very different is you don’t need much configuration in Warp. It works out of the box. Things like completions just ship with Warp, and we give you a nicer way of interacting with them. The prompt comes pre-configured. We also ship with a cool feature called Workflows, which lets you search for commands that are hard to remember.

For instance, the one that I always go back to is like, how do I undo my last git commit? That’s hard to actually figure out how to do via just completions. Typically, you’d have to leave the terminal and go StackOverflow and do a search.

In Warp, we ship with a community-sourced version of these commonly used commands. We also ship with an integration into OpenAI codex that does natural language command search. You can type in natural language what you want, e.g. “archive current directory”, and the feature returns the command for you to run, e.g. tar -cvf archive.tar *.

We have a really nice modern interface that includes a command palette, and we have built-in theming. You can to use it with your existing shell. It supports Bash, ZSH and Fish shells, but you immediately have a nicer experience working with the command line.  Users by and large are able to download Warp, open it, and immediately be more productive.

Is the tech stack exclusively Rust?

Our tech stack is almost about 98% Rust on the client and a little bit of Objective-C just to interface with the Mac platform. You have to create an Objective-C app. Our graphics code is in Metal, which is Apple’s graphics API. Our server is in Go. We have a small web app front end for some things, like the sharing of blocks, and that’s written in React and Typescript.

Does Rust interface directly with the Metal APIs or do you have to go through the Objective-C layer first?

You have to go through a very thin Objective-C binding where you basically pass in a shader program. The Metal APIs are in Objective-C, but it’s a very thin layer.

The most interesting thing in our usage of Rust is that we’ve built a full UI framework to do the rendering of Warp. We are not using one of the existing ones. That UI framework is something that we will open source in the future - it’s kind of inspired by Flutter. We had some early help from Nathan Sobo, the creator of Atom, on this library.

We set up the UI framework so that we can take advantage of Rust’s concurrency so we can lay out independent parts of our UI tree in parallel. We did a bunch of stuff with Rust concurrency in the way that the terminal itself is built. All of the things around updating the terminal model have their own thread. The UI is running on a completely different thread. If you have a ton of text that’s coming back from the Terminal, you can process that without janking up the UI. It was very hard to get the ownership rules to work properly for this in Rust.

We prototyped the app in Electron and it was kind of slow. We did that because we knew web tech the best, and it was easiest to prove out some of the product ideas. We then were looking for what’s going to be the thing that is, first and foremost, the best performance, but also, secondly, the best cross-platform story.

From a performance standpoint, what we like about Rust is that the binary compiles down to be as fast as if you had written it in C or C++. In some cases, even faster. We like the ergonomics of Rust. It has a nice set of libraries that are better to use than what you would get in C and C++. They’re more constrained.

The safety guarantees that we get with Rust are great, meaning we don’t have the risk of dangling pointers and random segfaults, and that kind of stuff. We do not have much in terms of crashes. Our more typical crashes happen with interfacing with the Objective-C code, which is the hardest part, but Rust has a nice foreign function interface.

A lot of the stuff that the Servo folks did at Mozilla was useful to us. It’s for doing the font rendering, text layout and Objective-C bindings.

Rust is a fast language. It has one of the best stories around binding into WebAssembly. We haven’t done that yet, but a major concern for us is that we want to have web rendering later, we don’t want to have to write it from scratch. Our plan is to take the same code base, and compile it to WebAssembly. It also has a good cross-platform story for things like Linux and Windows, which I think will be very useful for us when we support those platforms.

Rust also has a great community. If you look at the GitHub survey and what languages developers most want to work in these days, it’s Rust by a significant margin, which is cool.

How does this turn into a business?

We’re trying to build a company out of the terminal. The idea is to build something with a business model that’s much more similar to a company like Figma or Notion where Warp is free for individuals, and then there’s a set of paid team features that are focused on collaboration and sharing, and things like firefighting or making the terminal more secure for enterprises.

These are a set of features that hopefully large companies would one day pay for, but we don’t currently have a paid plan. We’re just focused on making developers as productive as possible by giving Warp away.

As far as our open source strategy, it is likely we will open source the entire Warp client but not the server side components. We are also aiming to create an ecosystem around well-defined APIs in Warp. For instance, if people want to add themes or they want to add workflows, we already have open source repos where folks can start contributing those, and they’ll go out to all Warp users. I see us adding more of those hooks and extension points into Warp as the product becomes more mature and we have more folks using it.

What does a “day in the life” look like for you?

I spend a lot of time thinking about and working on what the product should be like. I’m most interested in how we turn the existing terminal experience into something awesome.

I still code once a week. I think doing some coding is important and it’s a chance to use Warp. We dogfood our own product, so I get to see what the state of Warp is. When it comes to making technical decisions, architecture decisions, and understanding what engineers on the team are dealing with, it’s good to be in the code.

As a startup founder, I have to spend significant time recruiting and growing the team. We’re trying to grow at a measured pace. We are bringing on awesome engineers, and we are also hiring people who are focused on content and community: helping engineers understand what Warp is about and helping them contribute to Warp.

My day is a real mix, but I try to focus on where I can have the most impact to increase the chances of the company succeeding.

What is the team structure around Warp?

We’re a team of 14 right now. Alongside myself, there are 9 engineers on the team. We also have a designer, someone who runs all of our customer feedback channels and a chief of staff, who helps with internal operations of the company. We just hired our first Developer Advocate as well.

Our structure is flat. There’s no internal hierarchy. The way that engineers work on things is based on some combination of what they are most interested in, what does the company need most, and what they are most knowledgeable about. We rotate people through and let folks, as much as possible, choose what it is that they want to work on and give everyone a chance to lead different projects.

How did you first get into software development?

I did a little bit of programming when I was a kid, but I’m not one of those engineers who was heavily coding when he was 12 years old. I got into it in college, in a more serious way. I remember my freshman year CS class. I hadn’t planned on doing engineering, but I just got really into it and found it fun to build stuff. It wasn’t a straight line from college into engineering, It took me a while to figure out that I actually really wanted to build technology for my career.

After college, I worked at Amazon, then I worked at JPL, which is part of NASA, doing stuff that’s more like MATLAB type programming. I actually got a master’s degree in philosophy of science. My undergrad was symbolic systems, which is part computer science, part philosophy, part math. I did a lot of logic and I was really interested in that.

Afterwards I went to law school for a year, but I realized that this wasn’t the right thing for me. It was a hard choice to make at the time, but in retrospect was a really smart choice, because I enjoy what I’m doing now about a million times more than I would enjoy being a lawyer!

I worked in a music studio for a little bit. Then I found my first proper programming job on Craigslist because I needed to make some money on the side when I was working in the music studio. This was in 2006, and I got back into coding, and then, that turned into a job at Google, which is where I think I really learned how to program as a professional engineer.

I’ve been out of Google for a while now, and I’ve been learning how to be an entrepreneur and start companies and build products on my own.

What languages did you learn over the years?

I was coding C++ at Amazon. I learned MATLAB, which is its own thing. At Google, I primarily did Java and JavaScript. After Google I worked on an iOS app and learned Objective-C.

At Warp, we’re mostly Rust. So I’ve learned Rust and Go. I would say I’m mediocre at Go, but I’m pretty competent in Rust at this point. Rust is now my preferred language from an abstraction standpoint. I like the fact that when you use Rust by default, the thing you build runs really quickly, which was not the case for the web apps that I worked on, which you needed to put a lot of work into to get them to run with okay performance.

Tell me about your computer hardware setup

I’m working in New Mexico away from my main apartment so all I have is a 16 inch MacBook Pro M1!

Describe your computer software setup

OS: macOS.

Browser: Chrome.

Email: Gmail.

Chat: Slack.

IDE: VS Code.

Source control: GitHub.

Describe your desk setup

Nothing fancy. I do have this nice office chair that I have been using since I was at Google, called the Okamura Contessa. I’m a big fan of the Contessa chair.

The desk of Zach Lloyd, Warp

When coding

Daytime or nighttime? Daytime.

Tea or coffee? Coffee.

Silence or music? Music.

What non-tech activities do you like to do?

I go skiing every weekend during the winter. I also enjoy playing guitar and golf.

At the moment, I’m watching a lot of these startup failure dramedies now, such as WeCrashed and the Droput. Otherwise, I like watching football and to bet on it, through a Survivor Pool every year. I also like to hang out with my dog, Blue.

Find out more

Warp is a GPU accelerated terminal. It was featured as an “Interesting Tool” in the Console newsletter on 14 Apr 2022. This interview was conducted on 31 Mar 2022.