The future of the terminal is not a terminal | Warp’s Zach Lloyd

27:00And it's just proven to be a very natural interface for doing that type of like, you prompt the computer to do something type work. So that's kind of where we're at today. It's interesting how you describe it, how it evolves from the interrogative, asking questions to the imperative and making commands. And we've seen this evolution in how folks interact with AI and use it with tooling. And kind of what you're arriving at sounds so similar to me to what we would consider as to like a coding agent that lives in your in your terminal that's what it is today yeah so like what you describe it's like it makes me think of you know you might use like amazon q cli right or you might use claude code and you would or goose right and you would use it in the terminal but in this case the terminal itself is the agentic tool than that is what you built.

27:51And so as part of that, do you find yourself, are those competing tools or are they like a better together type of use case? They started off as sort of better together. Today, it's really competing. Like CloudCode, Ader, Goose, all of these terminal-based coding agents, they are like a, I would describe them as like doing a subset of what Warp can do. And I think Warp is actually the better interface to do what you do with those tools. The advantage that Warp has compared to something like CloudCode, which is a very cool tool, is that CloudCode is running as a CLI app. So it's something you boot up your terminal, you run Cloud, and then you're inside of a REPL inside of your terminal.

28:37and the downside of that is kind of like the downside that Warp itself is trying to address five years ago where if you're running an app inside of a terminal, you're really limited in terms of what you can do with the user experience. So you can only render characters. The mouse doesn't work again, which is annoying. You lose the ability to do any kind of richer UI. So like in Warp, for instance, when we present a diff to the user, they can edit the diff using their mouse and like WYSIWYG tools whereas when CloudCode presents a diff it's like an immutable thing that is just like printed to the screen.

29:20CloudCode you can like run multiple of them but because you're running them as like multiple terminal apps you you don't get like any kind of like layer of management for them at the actual app layer so like for warp for instance it's like we let you have like a a nice like panel that shows all of your running agents we give you system notifications when agents need your attention and so just by moving up one layer in the stack to be the actual like platform as opposed to an app running in the platform we can offer a a much better richer experience uh than one of those other apps and like for instance like another great example of this is so when you run a an arbitrary repl within warp so say you're running like the node repl or like the postgres repl warp can give you ai inside of those running apps uh by virtue of the fact that we are the terminal whereas like something like cloud code or goose or whatever they don't have any access to the platform layers it's like the difference between like being ios and being an ios app so i do think We're at a way more powerful layer in the stack.

30:31And so, yeah, we are essentially competing with those tools, I think, with a better experience at this point. It's cool how you're right, that it's the tool itself, so you get all the context baked in through all of the layers. Everyone knows when you use a terminal, it becomes inception. You're like application in and application in and application, but we're talking about the core of it, you know? It's kind of like, there's a lot of people who really love using Tmux. I don't know if you... I use Tmux. Yeah, I use Tmux. So Tmux drives me crazy because I'm like, it's like, why are we like emulating a terminal within the terminal as opposed to just building the features that you might like?

31:08I want panes. I want split windows and panes. Yeah, and like we just have that natively at like the app level. And that's just an improved terminal experience, right? Because I would use Tmux if it was built into the terminal I was using. Yes. And so in Warp it is, but people still... Like there are, I don't want to be totally dismissive. Like there's cool stuff that Tmux does that Warp doesn't currently do. And likewise, there are some advantages to just being a pure CLI app. I don't want to be dismissive of those tools either. Like, like you can run those tools like on, on CI or something like you don't, you don't need the UI.

31:47And so there's use cases where I think they're better. But for the most part, if what your workflow is as a developer today is like, hey, you want to come in, open up your tool and start multitasking with agents to fix bugs, to build features, to investigate production. I think it's way better to do it at the layer of like the outer app, not at like an app running within the terminal. Not nesting it inside of the team looks and splitting it all up and getting all of your henchmen doing your stuff completely. And what you're describing actually is such a critical part of building the terminal, I think, which is why it's so fascinating how you're tackling the problem.

32:29Because terminal is one of those things where folks get really opinionated about how their terminal can and should work. I'm like a.files disciple. I have a repo that I manage, right? And I have my team looks config. Like I have my opinions for how all of these things work. And so like when I'm going to use a new tool, I want it to play nicely with those things. But at the same time, I want to take advantage of new opportunities. I'm willing to throw away some of that personalization I've scraped on top of maybe. I've personalized it to make the tool that wasn't user-friendly, more user-friendly for me.

33:05But if a more user-friendly tool comes along, then that's an opportunity for me as a developer to try something new, which is like what you're building. You have an amazing attitude. Like there are I think that that's an awesome way of of of looking at it. Like there. Do you encounter the opposite from folks at the terminal? I do. On Hacker News, there are people who think that warp is akin to a sacrilegious thing because it breaks some of the assumptions that people have historically made between terminals and shells and.files and just how that whole ecosystem has traditionally worked. and that if you are a like terminal power user who has spent a lot of time building muscle memory or configuring that and then you come into warp and it doesn't work the way you expect, I think it's very understandable that you would be frustrated with that.

34:08On the flip side, my feeling is that setting all of that stuff up asks the user to do an awful lot of work and puts a very, very high barrier to entry in terms of getting like a good terminal user experience to the point where I think it really like turns a lot of people off of using the command line at all. And so, you know, and this is the camp that I was in and I'm not like a, like I'm a pretty experienced developer. Like I was a principal engineer at Google. I built a lot of Google Sheets product and like, but I was, I'm of the development mindset where I'm like a little bit lazier and I'm just like I want things that just work I want defaults that are sensible I want UX patterns that are familiar to me and I think a lot of the world is like that uh and I always felt like there was kind of a like a sort of vocal minority of power users who like the existing thing that made it harder to innovate and improve the base experience for the rest of developers and so you know I that's what we've tried to do but it definitely is like irritated some people who are maybe more traditionalists and that's okay well it's like there's a lot of opportunity to customize the stuff but I think what people have to remember is that we customize it ultimately because it doesn't it it doesn't quite hit what we need we need all these extra things so we do take the time as a developer to go in and sprinkle them in and then you end up like well if I'm sprinkling them in and I'm the only one who's using this, I'm going to make it perfectly tailored for me.

35:48And then this is where you get that like staunchly entrenched configurer, you know, dot files configured developer. But I'm definitely at the mindset where it's like trying out new tooling and new ways of using these tools is what we all need to do right now is what we talk about every week on Dev Interrupted. Everyone should be doing that. Everyone should be. And so it's like a huge opportunity. And like a lot of the, it's a huge opportunity. I agree with that completely. And a lot of the kind of older ways of working that were very like tuned. They were just tuned for a world that existed before AI.

36:22And I also think there's a lot of developers who are kind of tired of hearing about AI. But just to give an example, it's like you may have like some crazy list of aliases that you've set up around doing like complicated Git things. You may have like very complex like completions set up again for like you want your tab complete to do all this stuff. All that makes it's just like less necessary and less useful in a world where you can just express your intent directly in English to do is to do a task. And the AI will help do it for you. And there will be some people who like lament the loss of like, oh, but I used to know like the 50 different flags.

37:05but like think of the space it opens up in your brain and in your day to do what i would imagine is like more fun more cool like deeper problem solving and like it to the extent it moves you up a layer of abstraction and frees your mind from having to deal with that minutiae to me that's just so much better and even more fun as a developer uh and it's not about like replacing like it's it's like getting rid of the kind of work that I don't like doing as a developer like oh man I have these merge conflicts that's not a good use of my brain power a good use of my brain power is like let me think how the product should work let me think how the system should be architected I think like you have to at least try these new tools and try them with an open mind like what what you were saying absolutely and this kind of gets now at the core of our discussion I think about where this is all going and how the roles of developers will evolve because we are all moving up the cognitive food chain so to speak and what we're what we're doing what we focus on um in our day-to-day jobs is much broader and it's more decision management it's about understanding the product and the impact of what we're building uh building for the customer the end user as opposed to like you you like you just alluded to kind of getting lost in this like um sea of all the minutiae of trying to execute those obscure git commands or figure out, you know, all of these things that you typically have memorized or aliased to do your job when you can ultimately express those needs with natural language.

38:40So when we're stepping into this new world, I'm curious to know, where do you see this going for a developer as somebody who orchestrates agents and how is work building for that? Yeah, so that's that's what our entire product is oriented around right now is what i think the new workflow is going to be um i think the new workflow if you're a developer is you're gonna you're gonna have a bunch of stuff to build you're gonna have a bunch of stuff to fix that's not going to change but the way that you use the tools to build and to fix and to improve your systems and to ship product is going to change to one where it's like your one higher level of abstraction as you do your work.

39:25And so your typical day might be like, hey, you're going to come in, you're going to open up. Let's just say you're going to open up Warp for now because that's the product that I'm working on. You're going to have like things you want to accomplish. So you might want to build a future. And so you're going to start by sort of telling your computer, hey, I want to build this future. Here is the context. Here are the designs. I don't think right now you are typically going to like one shot it. Right. And I think this is like a fear that a lot of developers have is that it's going to be just like type of prompt walk away.

40:03Like I'm using these tools daily to do my, my normal development work and like they can one shot simple things. I think the complexity of the things that they will be able to one shot is going to go up, but really you're still going to have to be like, let me. work with, let me think about how to decompose this problem. Let me think about how to check in on the progress to make sure that, you know, if I'm going to delegate pieces of it to an AI, it's not going too far down the wrong path. But you're going to eventually get to a point where you start to kick something off and it could be like, hey, go up to point X and then check back in with me or check back in with me.

40:42If you hit some error that you can't resolve, you're going to start that task and then you're going to move on to your next task. And that task today as an agent task might run for like one to 10 minutes in a year from now. I think one of the axes that you're going to measure the progress of this stuff on is like how long it runs for. It might run for 10 minutes to 30 minutes. It might start running for an hour, but you're going to start another task and you're going to go through that same process of like, let me work with an agent to decompose this task in a way that makes sense let me understand what the agent is proposing and so i think this is one of the other failure modes as i as i work more and more closely with these tools like warp or cloud code or cursor i've worked with them all like a real failure mode if you're a real developer using these tools is like set them free and and you come back and there's like a thousand lines of code written and you're like i don't understand why it did this.

41:40We could never check this in. Like maybe it kind of works, but it's like, this was a waste of time. And so these tools are going to have to become really good at explaining their thought process to you and making sure that you as a developer are not just like doing dumb stuff. Like, you know, there's a very high risk of shipping bugs or shipping security issues. And so the like explanation and comprehension piece of it, I think is really important um but you're going to be working closely with these agents almost like they were like you know i don't like to anthropomorphize it too much but kind of like they're like like teammates a little bit uh and you're gonna have multiple of them going and the net effect to you as a developer is that it's going to be a a multiplier of what you're able to get done It just, it truly will be.

42:31It's going to make you do harder work. I really think in the sense of like, you're going to have to use your brain to do the stuff that the agent is not good at doing right now, which is like good abstractions, good system design, good performance, making sure code goes into the right place. But you're going to let the agent write a lot of your code. Like you're not going to be typing code very much. I don't think you're not going to be typing commands very much. You're going to be thinking more and like guiding more in English. And so I think that's the workflow that's coming. And so then as a someone who's building tools in this space, what we are trying to do is build the right set of tooling to support that workflow.

43:23and I don't like I said earlier in this conversation I don't think that tooling exists I don't think it's the IDE I don't think it's the terminal I don't really think it's something like Devin or like these things that are like autonomously running around your code base not today at least maybe maybe those work for very simple tasks but because you need the developer's brain in this deeper way I think it's the tool that like best allows the developer to do kind of the hard thinking and the orchestration to get the results they want. Yeah, you're touching on the skills, too, that developers will need to embody to be better developers.

44:04And we've heard this from a lot of folks on Dev Interrupted, Dr. Ashoori from IBM, who talked about the new skills that developers will need to have. And really, it focused around you will become a decision manager more than an action doer, which really aligns with what you're talking about in terms of setting these agents off to do their tasks. And you're not really writing the code. You're understanding the abstractions and the end result of what you're building, right? So it's like a new skill set, really. It really can be night and day from some of the skill sets that traditional engineers or engineers until now have been building.

44:39Are there other skills in that toolbox that you think are really critical for folks to start building as they start kind of moving up the food chain? there's the foundational stuff that you you get in when you study computer science or you study math or basically study anything with like a like rigorous reasoning i think like problem solving is your number one skill still as a developer it's just like it's like where in the where in the stack are you doing that like back in the day you know you were you were doing it with assembler and then along came like c and then came c plus plus and then came java and then python and then javascript and like the the distance from the metal increased but the fundamental like skills of like logic and problem solving and just like reasoning through tasks that's a constant thing and i i don't see any world where that goes away and if anything that's like more and more important.

45:45Can you see the big picture? Can you see how pieces fit together? So that's like big, big picture way more tactically. Um, I think like you're going to have to learn how to use this new tool set. And that's like, again, I don't want to overemphasize that. Cause I don't think like a developer is like defined by how well they can like use their IDE or the terminal or something but there is an actual skill set in in how do you how do you prompt how do you decompose problems in a way that that the models are good at doing bite-sized pieces of them this is all like totally new territory uh yeah where it's like there's no real best practices no one is learning how to do this but you know kind of i i would imagine that similar to like taking like a high level programming class at some point you're going to take a high level class or course on like how to effectively work with ai to build systems and so you know a lot of people miss that i think like a lot of good engineers will have their first experience with ai be something like they hit some issue and they're like okay i want you to fix this thing for me and like they'll literally prompt it like that and like that's not going to get the job done any more than like, you know, like you have to explain in a lot of detail what it is that you want done in order for it to do it.

47:18It can't read your mind. It's not like magic. And having that kind of bad experience, I think has led people to believe that the technology is, is useless, or it's like, it's not there yet. And like, there's some truth to that. You know, even I think Like if you tried, you know, working closely with AI six months ago, you would find that like, hey, it's not quite ready. But the rate of progress is like so fast that, you know, having one bad experience like that is like you have to get past it. You have to learn how to use the new tools. And, you know, you have to kind of do like a continuous education thing because the power of the tools is changing very fast.

47:58What you're touching on is so true that folks who use these kinds of tools, maybe like six months ago and didn't have a great experience, never revisited them. Their mindset about what they can and can't do is fundamentally just, it's an era behind, it's a decade behind in terms of how far. It's crazy. The rate of progress is remarkable. The rate of progress is remarkable. And then the need for continuous education is just profound. Like, it's just like you need to be a voracious learner to be a developer now. And, you know, what advice would you give to one of those folks who maybe tried one of those tools and have a good experience and they're just like, you know, maybe they're a skeptic now or they don't know exactly how to get started.

48:40What would you say to get them back into the mix and trying out a tool like Warp? Yeah, so first off, it would just be like an open-mindedness, I think is really important. Approach it like a tool. so like i think that a thing that turns a lot of developers off to this stuff is it's overhyped it's like you see all these twitter demos of it doing stuff and they're like yeah right um and so so view it as a tool is my first is my first uh thing and so and what does it mean to view it as tool it means that there's like some um responsibility of the user of a tool to learn how to use it well so it's like you know you wouldn't like open up vim and be like well maybe you would like i i would open up and be like what the heck is this but like what is this how do i get how do i exit how do i close this or or get or whatever it's like there's a learning curve and so approach it from the standpoint of like hey um i'm gonna try this i'm gonna need to learn how to use it um i would say realize that it's a kind of weird tool in that it's like got like a sort of stochastic quality or like a random quality to it it's unlike any other tool that you use it's like purely deterministic whereas ai is is just not fundamentally there's like a I mean it is fundamentally a function I guess but like you have to use it as though as though there's like an element of randomness to it which is kind of a foreign experience for most for most developers yeah I would recommend like seeing if there are people on your team who are using it successfully and like learning how they're using it and having them share prompts that they're having success with, share tasks, share how they, like, what their workflow is.

50:46There is a lot of good content online on this. Like, I even, I'll plug, I do a live stream where I try to use AI to do real coding problems, not, like, like, one-shot marketing websites, which I think is, like, a lot of, like, what the people have seen. Yeah, make. Yeah. Yeah. And so I think seeing that type of development where it's like, I'll be like, okay, I'm going to go one step here. I'm going to have it right tests. I'm going to verify that the API looks good. Then I'm going to go to the next step. And so showing how you can use it to break down a problem. That would be my, my advice. And then you just like recognize that the stuff is changing super duper quickly.

51:32And so, you know, you may have tried it and, and it's, it's, it's really much better now. and will probably be much better six months from now. Yeah, exactly. And I love too that you learn in the open that you have a live stream where you kind of hack around with these tools. People can get a glimpse of your thought process, which I think is most critical, almost like the body doubling of, I'm going to watch another developer use this tool and it's going to critically fail for this guy and then what does he do to fix it? And it fails for me all the time. All the time, yeah. That's okay. Like at the end of the day, I still think it's like a net gain when used properly.

52:12You know, as we start to kind of wrap up our conversation, we've covered a lot of ground about what you've been disrupting, what you've been building, the skill set of developers of tomorrow. I'm wondering from your perspective, are there any kind of like top level points that you'd want to end on? If talking about like how developers will orchestrate with agents of tomorrow, perhaps there's something that you see about the future of development that you think a lot of your peers are missing or are maybe not quite on your level in terms of understanding where it's going to go? Any kind of points like that that you'd like to end it on?

52:46I think one big point that is maybe overlooked is like, this is a real change in behavior that we're asking developers to make. I think there's a lot of like, thought of like, okay, you put these capabilities out there, you do demos online, that developers will just overnight change the way that they work. Most developers are not working with agents right now. The extent of most developers' use of AI right now is autocomplete, which is a great feature, but I think it's fundamentally going to become obsolete as developers move to a way of working where they're not doing so much stuff by hand. They're not editing or coding by hand.

53:40And so from a product standpoint, what I think is really interesting to explore is like, how do you meet developers where they currently are in their current workflows? So whether it's in the terminal or the IDE and like get them without beating them over the head, like you need to work with AI, you need to prompt but get them to want to do it and get them to have that experience of like an aha moment that this is actually fun that this is actually like a cool new way of solving problems that this isn't like a thing that's competitive with me um because i feel like there's a lot of like this is going to take my job this is going to like replace me i have just the opposite experience at this point, which is like, this is super fun.

54:28This is like a, I'm playing a game of like trying to get this like alien brain to like help me do my work. And it's like a great puzzle to solve, which is like one of the things that I love about coding to begin with and development to begin with is that puzzle solving nature. And so, you know, I'm, I'm just trying to think for warp. This is a thing that's very top of mind for us of like, how do we get developers to have that experience of like this doesn't suck this is actually fun this lets me get more done this lets me focus on like stuff that i want to do and get them to experience that as opposed to be told that they need to start working in that way so i think whoever's going to be really successful um in this new like product space is going to be uh a product that like gets really good at that because because that changing developers behavior is really hard.

55:23And I know that from being in this space for a while now. And the best way to change their behavior is to just like give them the experience of the new thing and then have them want to change it. So that's my last thought. Amazing. Well, that's really well said, Zach. And, you know, we've covered a lot of ground here about how warp is evolving to meet those new needs of the developer and kind of re-envisioning the whole developer experience. And I can't think of a better note to end on than to say that, you know, you have to get folks excited about where this is going and you have to build that internal momentum, not by saying this is how it must be, but seeing this is how it can be.

56:01Exactly. Showing a better world. And I think that you said that really well. And for folks that are curious about like what we talked about today, where can they go to learn more about Warp and what you're building? Just go to warp.dev. You can download it. It's totally free to get started. You can watch videos on it. you can check out our YouTube channel if you want to like see us doing interesting stuff and learn more about the product before you try it but yeah that's it Amazing well then we'll drop those links in the show notes so folks can check it out and to you if you've been listening you know this conversation is one part of a long saga conversation we've been having on Dev Interrupted about how the developer experience is evolving how people are using new tools to get more work done so if this is something that is really interesting to you that you're working on on your teams Zach and I want to hear about it.

56:51So come on LinkedIn and tag us about today's episode. We're posting this episode on Substack and LinkedIn. It's really easy to give us a notification and let us know what your team is doing that was successful or unsuccessful or what your own personal experiences were around using these kinds of tools. Because when we're all having conversations about it, we can all get better informed about where it's going. So thanks for joining us for this conversation. and we'll see you next time on Dev Interrupted. Thanks for having me. This was awesome.