In this RedMonk Conversation, Rachel Stephens sits down with Zach Lloyd, CEO of Warp, to unpack how the terminal has evolved from an overlooked productivity bottleneck into the center of the agentic development workflow. Zach traces Warp’s journey from reimagining the terminal for broader developer accessibility to redefining it as an “agentic development environment” (ADE), designed around prompting, reviewing, and collaborating with AI agents rather than hand-editing code. The conversation explores why the terminal is uniquely positioned in the software stack, how Warp blends elements of terminals and IDEs without being either, and what this shift means for collaboration, context management and automation.
Links
Transcript
Rachel Stephens (00:12)
Hi, welcome to RedMonk Conversations. I’m Rachel Stephens and with me today is Zach Lloyd from Warp. Zach is the CEO of Warp. It’s a company that I first heard about back in 2022. And at the time they were thinking about the product in terms of re-imagining the terminal experience. In particular, trying to make it easier for a broader group of developers to use. So it wasn’t just a tool for power users, but it was something that everyone could be more effective with and made it easier to learn, less of a steep learning curve.
I was impressed at the time and since then the market in DevTools has shifted so very substantially that it has pivoted, I don’t know if pivot’s the right word, but it’s definitely expanded its scope in some really interesting ways. Zach, I’m so excited to have you here. Tell me about your journey building
Zach Lloyd (00:58)
Yeah, so I’m Zach. My background, I’ve been an engineer for a really long time. The mission of Warp has always been the same. It’s to empower developers to ship better software more quickly. Your sort of preamble on Warp is pretty accurate. we set out with a product vision of trying to reimagine the terminal with the idea being like this is a everyday tool for most developers.
It’s something that from like a product experience, I don’t like very much. Like, just think the stock terminal people are wasting a ton of time, experiencing a lot of frustration in terms of like how hard it is to use. And conversely, if you get good at using it, you can get a lot more done. the current version of warp is quite, is like similar and different. Like we, basically have taken all the work that we did to make it an incredible terminal experience and made it work for agentic development. And so.
We don’t even call warp a terminal at this point. It’s really, we call it an agentic development environment. And the idea is like, you can have a single interface for telling your computer what to do. You can tell it to do that in terminal commands if you want, and that works really well in warp, or you can tell it in natural language, in English. And if you tell it in English, then warp launches an agent to achieve that task.
And because of like where the terminal sits in the development stack, can do almost any kind of tasks. So Warp’s really, really good at coding. We’re one of the top agents on Sweet Bench and Terminal Bench, but we’re also really good at like setting up projects or debugging in production or doing stuff with Docker or your clouds. And so it’s like a really, really versatile, powerful spot in the stack to be. And like you said, that’s like transformed our business, which has been awesome.
I would say we were like kind of half right in our original thesis and just like, you know, we, I went into this being like the terminal is like an overlooked piece of real estate. It’s kind of like how I would imagine it. And like our original business model was going to be building collaboration features on top of that. And like that kind of sort of worked actually. And like, we have traction with that, but the, you know, we got pretty lucky. Like this transformational technology came along like, you know, a couple of years into the business, which applies super well.
to what we’re working on. now, again, we couldn’t have really predicted this, but the terminal is sort of like the center of the agentic development universe. Things like Claude code or Gemini CLI or warp are like the preferred interface for doing agentic development. So it’s been really good for us.
Rachel Stephens (03:16)
Mm-hmm.
Yeah, I have so many questions. So when you were originally thinking about collaborative terminal, that was like helping team members basically, if you have terminal commands, you can share them across your team. Was that the original version?
Zach Lloyd (03:36)
Yeah, the easiest way I can see of it is like Postman for your terminal. So Postman is like this API collaboration platform. A lot of that same type of thing applies in the terminal where it’s like, you might be executing complicated commands or you might want to put your on-call playbooks in a place where you can get to from the terminal. We even built like real-time, you know, multiplayer in the terminal, which is very cool, but like not that frequently used. So that was like…
Rachel Stephens (03:40)
Okay.
Okay.
Yeah.
Zach Lloyd (04:04)
And that kind of goes to my background. Although I was principal engineer at Google, I used to run engineering on the Google Docs Suite and I built a lot of Google Sheets. so, as a founder, think you kind of like do what you know. And that was a thing I knew how to do. But I think we’ve been pretty flexible and adapted well to the fact that market has changed. overall, it’s been a really positive change in the developer tools market. It’s gone from like…
A market where it’s really hard to make money, where it’s like you have some of the fastest growing startups in history are in this market now, which is cool.
Rachel Stephens (04:35)
Absolutely. It’s been a profound shift in the I won’t pay for any developer tools ever, too. I will pay $200 a month for some developer tools. It’s been…
Zach Lloyd (04:41)
Well, so
here’s the thing. It’s like, don’t even think of the market as like developer tools. I think that’s the wrong framing for it. It’s like, I think it’s the market for automating the production of software. that’s like, that’s where this is going. Because like,
Rachel Stephens (04:52)
Okay, how do you think about it?
That’s fair. Okay, yeah.
Mm-hmm.
Zach Lloyd (05:03)
Like, look, there is a developer tools aspect in terms of like a lot of what these tools do, Warp included, is they help your developers be more productive. But I think the really, really big opportunity is like, if you look at like the market for the production of software, and that could be enterprise software, consumer software. And I think the more interesting part is enterprise, honestly, not like the long tail of like five coded websites, but like…
Rachel Stephens (05:17)
Mm-hmm.
Zach Lloyd (05:26)
That’s a huge, huge cost to companies and also like a huge economic driver. And we’re getting to the point where big pieces of that can be, you know, can be automated. Whether it’s like, you know, you still need developers in the loop. So it’s not like total automation, but pieces of it are going to be totally automated. And that’s just like a gigantic economic opportunity.
Rachel Stephens (05:48)
Absolutely. So when we’re thinking about where you envisioned collaboration kind of being human to human collaboration, team collaboration, how did that kind of translate into building agentic tooling? Because it’s still collaboration, but like in a very different flavor of collaboration. Like, are there parallels or no?
Zach Lloyd (05:54)
Yep. Yeah, so actually the pieces, there are like, so the thing that we built on the collaboration side, which we call warp drive is like this knowledge store. And the original concept was like, you and your teammates can put organizational knowledge into it. That’s useful for using the terminal and like now all that knowledge is just available to our agents. So that’s, that’s like one very obvious thing, which is really cool.
So if you have this corpus of commands or notebooks or environment variables, or these days like shared prompts or whatever, the agent now, like we set it up so you can do rag over that. And the agent just understands your context way better. So we have that in addition to MCP and just like having the ability also to like share MCP, we have the ability to like share rules. So all of these things that make agents not just work for individuals and work for companies as part of our collaboration story, but.
I think what you’re asking is like, you actually as a developer now, the way you develop software is you kind of collaborate with an agent and the stuff that’s actually more useful for that that we’ve built has to do with like how we change the user experience in the terminal itself. And so in warp, like we have the terminal session isn’t just like a stream of text. We’ve essentially made it something that’s much more like.
you know, like a chat experience where it’s like the user says something, the agent responds, the agent does tool calls. Those tool calls can actually have like rich renderings, meaning like you can see like an actual diff. If you want to go and edit that diff, you can edit it. And so just the fact that we put all this effort into making the terminal interface, something that we own gives us this huge advantage when it comes to other tools that run in the terminal, which are just like,
plain old text-based apps and you can’t do any of these things with them. So that’s what I think a very differentiated offering in this space compared to everything else out there, which is either like a VS Code fork or it’s a pure text-based app and warp is like, no, we’re like an actual like native GUI app that has like some of the best parts of the terminal. Like it has that interface, that form factor, but we can also do things like code editing or code review or whatever we want with the UI and the terminal.
Rachel Stephens (08:11)
So that sounds kind of like a lightweight IDE embedded in the terminal. Is that their way to think about it?
Zach Lloyd (08:16)
So we call it an ADE. We are attempting to create this category of a Gentic development environment where our position on this is like neither the IDE nor the regular terminal is the right interface for doing a Gentic development. The way that development is happening is fundamentally changed. And what you want is something that is designed from the ground up to support that workflow. And like it’s going to have aspects of the terminal for sure. Like, but it also has aspects of the IDE.
Rachel Stephens (08:18)
Okay.
Zach Lloyd (08:43)
but it’s not an IDE because in the IDE, like the primary interface that you have looks like Microsoft Word for editing your coding files. And like it’s really geared towards hand editing. I mean, no shade to Microsoft Word, like hand editing, handwriting code is not the primary thing that developers are going to be doing. So having an entire interface dedicated to that doesn’t make sense. It’s like the secondary action. And so the primary action should be prompting an agent.
Rachel Stephens (09:05)
Right.
Zach Lloyd (09:11)
I think like the secondary actions that you really want to support are like reviewing the agent’s code. So actually if you’re like, I think code review becomes much more important and not like code review, in GitHub or whatever, but code review in the inner loop as the agent is writing code. And then if you need to having the ability to go and edit that code, but that’s, like I said, that’s, that should be your fallback. That shouldn’t be what the primary interface is geared towards.
Rachel Stephens (09:33)
Interesting. So when you’re thinking about your competitive set then, are you kind of thinking more of like the AI CLI tools or is it like the AI kind of IDE tools or is it a mix of both? It’s all of them?
Zach Lloyd (09:44)
It’s all of them because like, it’s like whatever tool someone’s adopting to build software by prompt or to build software automatically. And you know, there’s different interface approaches to it. Like there are the AI CLIs like cloud code and there’s the AI IDEs like cursor. I would just think of the competitive set in terms of like what you can do with the tool, not what the interface is. And so in warp, can, you can code. Like I said, it’s really, really good coding.
Rachel Stephens (09:45)
Okay.
Zach Lloyd (10:09)
We also, again, one thing I think we’re much better at or better positioned for than like the AI IDEs is doing all this other stuff that goes on in the software lifecycle, figuring out why a server is crashing or like setting up Terraform or like, you know, interacting with your AWS, you would never want to do that through an IDE. So that is our competitive set, but we are like, in a sense, even broader in terms of what functions you can do beyond just coding.
Rachel Stephens (10:31)
Yeah. Interesting. I like the concept of what the code can or what the tool can do versus the form factor of the tool. I think that’s going to be increasingly shifting in the next little bit. So talk to me about how you are building context for warp. I know you mentioned that you can do rag over warp drive, but like what are the primary places in which warp is kind of deriving all of its information?
Zach Lloyd (10:40)
Yeah, so there’s a lot of them. there’s code-based context is one thing I think is important. Actually, not every tool invests in this, but we vectorize the code bases and do embeddings on them and have empirically found that that actually increases efficiency by finding things more quickly in search for some use cases than just like grabbing over the code. So code-based context is one. Another big one is just like…
The fact that we’re in the CLI and CLI tool calling is such a wide surface of things that you can bring into the context window by executing a CLI command. So that’s probably our number one way of doing it is like, OK, if you want to understand your G Cloud setup, just have warp run a G Cloud command and ask about the servers. So there’s all this context gathering on demand for that. There is MCP. So we’re, you know,
fully supporting MCP and actually trying to make the MCP experience and work really nice and work not just for individuals but work for teams. Because MCP has this activation energy which is annoying. There’s stuff like rules, which is another big one. So you can do that on a per project basis or you can do it through our collaboration platform and get it globally. And that’s like, that kind of gets you most of everything.
You know, like through those tools, like again, the warp is super flexible. You can do web searches, you can attach screenshots. So like that’s other stuff that we’re looking at too is just like, do you, what are all the tools that you can bring in context? But it’s like a pretty rich set of things that you can bring into the agent. And then the other thing that we do a bunch of work on is like, with all that context, how do you present it efficiently so that the agent is looking at just the right stuff at the right time?
And there’s a lot of work that goes into the harness there, whether it’s summarizing context, doing subagents, truncating long context, using different models for different types of things. So that’s another place where we spend a lot of time working.
Rachel Stephens (12:50)
Interesting. there’s so many ways we could go with that, I do, I don’t want to do, I don’t want to derail us too much. I’m curious how you’re thinking about the vision of just agendic development overall. What do you see coming together here?
Zach Lloyd (12:53)
Yep. Go for wherever you want.
Yeah. So I think that there’s like three phases of development, roughly is how I see it. So there’s what I’ve done for most of my career, which I conceive of as developing by hand and that like in that flow, I’m like opening a code editor and typing code and then opening a terminal and running commands. I think we’re transitioning from that into developing by prompt. and like, I think we’re really early into this, despite all the hype, like the, the,
use cases where development by prompts is most successful for is what you would think of as like vibe coding. And it’s really like, you know, building websites or web apps. It’s kind of like the successor to Squarespace or WordPress, in my opinion. And like for professional context, we’re still very early in agentic development, but in agentic development, it’s like, you’re opening a tool like warp. are as a developer, telling the computer what you want to do. And that could be building a feature or fixing a bug.
And then I think there’s the third phase, is like automated development. And we’re interested in that as well. With that, like there’s different approaches. So there’s like the approach that we’re doing, which is basically trying to provide developers with programmable versions of our agents so that they can, you know, have the agent do some amount of their, their work on their own. that would be like, like we just did a hackathon in this. And for instance,
having an agent that automatically updates your documentation based on every code change is like, think a canonical example. And so we’re providing a version of warp where you don’t have to sit there in front of warp, typing into it and like watch it run locally. Instead, you can program it to handle triggers or run on crons or respond to system events and do some portion of either code writing or just honestly, it could be any like developer task.
Rachel Stephens (14:40)
Thanks
Zach Lloyd (14:45)
All of the stuff that we have in Warp’s coding agent is really useful. So yeah, those are the three phases by hand. It’s these interactive agents and then it’s like automating parts of the development workflow using agents.
Rachel Stephens (14:55)
And so like the concept of the asynchronous background running agent I think is compelling to me because in a lot of my experiences trying to play with agent workflows Like I end up kind of being like the XKCD comic about the compiler It’s like I’m just waiting for the agent to run better go have like a sword fight in the hall It’s I don’t want to lose my flow But I also like so I end up just waiting and that is clearly not the efficient way to do it So it’s like is that we’ve got to keep moving to that that next phase. It really is
Zach Lloyd (14:59)
Yeah.
Yeah.
Totally.
It’s the new compiling.
Rachel Stephens (15:25)
So for me, it’s fun to just kind of hear people’s visions on how this all comes together, because I think, as you said, we’re still really early in this phase of trying to figure out how all these workflows get incorporated.
Zach Lloyd (15:35)
Totally, yeah. from a business perspective, think automation is the best. It’s just like productivity improvements are hard to measure, but if you could just be like, sell automations or sell a piece of developer infrastructure where they can build these automations, which is how we’re thinking about it at Warp, then it’s more outcome oriented and you can charge in a different way. And it’s just easier to measure the value. So I do think that’s where it’s going.
Rachel Stephens (15:57)
Can you talk to us a little bit about your pricing? Cause I know you’re not doing tokens, right?
Zach Lloyd (16:01)
We do a credit system. the pricing for all these tools is very complicated, like for us, get, there’s our basic pricing plan as you buy a fixed number of credits and that recurs every month and that’s a subscription. And if you exhaust the number of credits, have the option to go onto a pay as you go model where you can just like buy credits all the cart.
Rachel Stephens (16:02)
Okay.
It really is.
Zach Lloyd (16:26)
That’s okay. It’s like, I don’t know the right, like I would be lying if I said exactly, I knew the right way to do this. Token pricing I’m like a little bit against from a business model perspective, just because it makes it very hard for us as we improve the efficiency of the model to capture any of those gains. If we’re just like, if it’s just like raw tokens that we’re charging for. Like if we do something, like I think we should be incentivized as a company to make the interactions more token efficient.
And if we’re just like passing through token costs or not. And so that that’s the reason I don’t like that model very much. I do think some sort of consumption based pricing is like pretty important. Otherwise it’s just like, what’s that? Yeah. You just end up spending so much money. It’s very, it’s very hard to price the fixed request plans.
Rachel Stephens (17:05)
Otherwise you get eaten alive. Yes. Absolutely.
Zach Lloyd (17:15)
in a way that’s incentive aligned without some sort of consumption based pricing, because otherwise like we end up just like losing more money, the more money, the more people use our AI and what we want to be in a position is like, we make more money as they use it more. So that’s like a consumption model. But yeah, it’s like, it’s messy for sure.
Rachel Stephens (17:19)
Thank I just I love hearing people’s general philosophies on all these things where we’re going how we price it like all of these things are all still so up in the air right now that it’s just interesting to hear how people are approaching it and what they’re finding works and doesn’t Wonderful well is there anything I didn’t think to ask about so like Like I don’t want to make you pitch but like if somebody is looking at warp Versus like a cursor like why would someone want to use warp versus like one of the AI tools that they’re already using What’s their switch point?
Zach Lloyd (17:43)
Totally.
Yeah, so great question. the, way that we’re trying to differentiate is really on the developer experience, meaning like when you go in to try to build something with agents, we want warp to be the easiest to use. We want our agents to be the highest quality. We want to give you the features that really let you see what the agent is doing and control it and review it. We’re really focused on working on real code bases. Like we’re not.
positioning ourselves as a vibe coding tool. And so if you like using Claude code or Gemini CLR, if you like that way of working, you should strongly consider a warp because it’s like, it’s a similar style of working, but with what I consider a much better developer experience. And like similar, it’s not more expensive really. It’s like maybe like it’s in the, it’s comparable cost wise, but just like, from a product and usability standpoint, that’s like my DNA, what I really care about. And so when you come in, you should just be able to get more done and have like more fun and doing it and work.
Rachel Stephens (19:01)
Well, Zach, thank you so much for carving time out of what I’m sure is a very busy schedule as a CEO to talk to me. But I appreciate it, though. This has been great. And if you have anyone who wants to look up Warp or look up you, where should they go?
Zach Lloyd (19:05)
I’m very busy. Yeah, but this is fine. This is cool. For sure. Just go to warp.dev. You don’t need to look me up. If you want, I’m on LinkedIn. I’m writing my thoughts on LinkedIn, that’s like the social channel that I’m most on. I’m not very good at Twitter, but just go to warp.dev. Check out the product. if you’re, we have a preview built too for people who want to get the early access to stuff. So, but just go check it out.
Rachel Stephens (19:20)
Okay, fair enough.
Thank you very much. Have a good day.
Zach Lloyd (19:38)
Thanks, Rachel.
No Comments