The End of Specialization? How AI Shapes Modern Dev Teams | Vercel’s Lee Robinson

26:17And you might not have ever written a line of code before. And that's fine. So all of that context to say the makeup of the teams is then shifting to where you see more and more people who maybe focus a bit less on the code as their full-time effort. And they're more of an architect or a planner or a product thinker. They're much closer to the customer experience. They're much closer to the product experience on delivering actual value to people who are giving them money. And I think over the long term, it will change even further the way these teams operate. There's so much good stuff in there.

26:56I want to unpack some of it because you covered a lot of really great points that I really identify with that I'm also seeing as well when I talk with people, when I interact with folks that are adjacent to tech, especially and how they approach the tools and what they can create. It's actually quite profound what someone can build with little to no technical knowledge. And then what impresses me even more as like a former teacher, like being in a classroom is I was like, wow, it's really hard to, you know, I maybe couldn't teach my grandma Python very well, but I could probably teach my grandma what makes a good product and what makes a bad one and how to iterate through it, right?

27:30She can cook a recipe in the same way that I could explain to her how you make something good. And that's a huge unlock, right? It's like now suddenly you have this whole field of, we call them whatever you Some people are calling like citizen developers. I think it's just like the new norm of access to this technology, of being able to go, there's something like you described of like V0, type in a concept and get it back. I liked how you framed it. I like how you defined it as personal software. The term I've been using, and we've been kind of talking about a bit on Dev Interrupted, is disposable software.

28:00Of like, you just almost need it for a one-off purpose. Or the idea of something before that like, oh, you know, you do it once and it's a little annoying. you do it twice and now you're like repeating yourself do it three times now i need to automate it well maybe this would be what you build around like time number two you're like i did it the first time i kind of knew what i need if i could just build this little eject like tool or toy to like just do it when i need it on demand now i don't have to think about like automating it maybe this isn't worth automating um i'm wondering too about how do you see teams and stuff take advantage of things like automation to improve their productivity and their developer experience because i can't think of something that's has a faster shipping timeline that has a better workflows and pipelines than web dev, which has always been on the front line of those kinds of tools.

28:47So like, what kind of things are you seeing? Yeah, I think for a lot of the people who just have an idea, but they don't know how to build it. There's always been this like long running meme or joke of you have that friend who's the idea guy and they're like, I have an app. I need, I need a guy to build my app? Like, can you build my app? Yes, you took this right out of my mouth. I think the same joke. It's so good. Yeah, I think that for I believe the kids are calling them normies, or the general population for most people. Yeah, you know, they go to the app store and they download apps. They use a lot of websites, too.

29:24But I think a lot of people are exposed to their phones and downloading an app on their phone, they download Instagram or whatever. And for these people, they think, and I have this idea, I want to build an app, or I want to build a product based on that. They don't really know exactly how to go from A to B, like draw the rest of the owl meme. Like, okay, I think I know what I need to do, but I need to talk to somebody to actually get it done. And I think what these tools are doing is they're allowing people to actually take that first step and just try something out. And maybe it's not perfect.

30:02Maybe it's, you know, still needs some work, but they've got something that they can then iterate and build from. I know at least for myself, whenever I'm learning something new, it's the, like, when you know there's this really complex and difficult and daunting task, it's very easy to get stuck in knowing I've got a lot of stuff to do. How am I actually going to get to the end? And the only way for me personally is like making very small incremental steps and then having some kind of reward system along the way where I'm like, oh, wow. Okay. Awesome. Like I did it. I see it. It works. And as a personal anecdote, when I was learning to program somewhere along the way, I started to get into app development because I thought this seems really interesting.

30:45I'll learn how to do it. and the speed from idea to published app was just so incredibly slow that it really just took all the wind out of my sails for wanting to build app development. Not only was it very hard for me to come up with ideas and then turn it into code and iterate on the idea and then actually get it distributed on the app store, but then once it got out there, I was like beholden to the strange Apple rules. And if you made a mistake, now you had to push a new version and that version had to get updated. And it was like, oh my gosh, this is a lot. And it really made me like re-fall in love with the simplicity of the open web.

31:31Anybody can just publish stuff. You don't need permission from anybody. Anyone around the world can buy a server and put their site online. They can upload files to a CDN and distribute it around the world. They can use a service like Bracell. It's got a free tier and get a site online. They can use a tool like V0 or others and just type in their idea, click a button to deploy to Bracell and it's live. And I think that style of tool will really democratize the power of the open web, which is that anybody can build any idea without having to ask anybody. That's what gets me really excited, I think, about this next generation of tools.

32:07You really couldn't have said it better. I also really like the idea of now the ideas guy, and they have everything they need to execute their idea. And I'm just going to say the idea guy, like, a link to v0. Yeah, it's his dream come true. And so along the way, you know, there are those obvious wins about the building. But also when teams are working and they're deploying, you called out a really good key differentiator of the open web. But what I like about it, too, is that anyone can build on it. You can get on there fast. It's not a closed garden. You don't have to be beholden to someone else.

32:40And so that really levels the playing field. But that also allows for really rapid iteration. You know, you pointed out how there's like so much tooling and so much opportunity available specifically to take advantage of these AI tools on an interface like the web. Because it's evolving at light speed. It's evolving as fast as people can upload stuff to it. So along the way, the fundamentals become even more important because now you've built this one-off app and you've ejected it and you have your proof of concept, as you said. You're tinkering with it. It works. But now you need to take it from 80 to 90.

33:15And then you need to take it from 90 to 100. And those are two very different leaps and bounds in the development process. and for hobbyists or citizen developers, you know, they've never taken anything from 80 to 90 to 90 to 100. That's a whole new territory for them. They've been focused on zero to one this whole time. And so when you talk about that, what kind of opportunities and gains do you see for teams that are taking things from 80 to 90, 90 to 100, 99 to 100, you know? Are there specific kind of tools or orchestration or ways that teams like talk about what they're doing week to week that you see really kind of like shine in terms of high performing teams?

33:55Yeah, so my personal experience is that AI tools and especially around code generation are very helpful from going from zero to one or from zero to 50 % in your analogy from the 80 % to 100%. It's a little more difficult because at that point, you need to have more of an opinion about what exactly you're trying to build than what you need to get done. At that point, it's more of a consultant that you're asking questions to versus the actual implementer of the idea. So I think especially for larger teams who are working on more of the 80 to 90%, it's mostly done. We just have to like add the polish on top.

34:40The biggest unlock of AI, honestly, is just automating the simple repetitive things that maybe would have taken an hour that now you can get done in five minutes. And then you free up that time to go back and do other work that is furthering actually building the product. One of the things that we talk about with VZero a lot is this idea of code last. So different than a no code or low code tool. Code last is, yes, we still care about the code. We still want to have high quality code that works and compiles and uses the latest tools and is accessible and performant. it's just not the first and foremost thing that you're concerned about when you're trying to build a great product.

35:21It's make it work, make it right, and then make it fast. And making it work is a lot of iterations first. So I think if you apply that same kind of code last mentality to the entire software development lifecycle, yes, you're still going to write code. It's still going to have code review. You're still going to work together on teams. Like none of that stuff really goes away. It's just your relationship with the code is changing a bit more. Yeah. Instead of it being the thing you have to do first and then really lay down a thick foundation with, it's now just another ingredient. Right. And there's also different, you know, caveats always apply.

35:58If you're working on mission critical production software that handles payments, like you're probably not vibe coding out AI generated code there. You're probably being much more particular about the code that gets shipped, you have multiple reviews, you have lots of processes in place, you have testing, you have multiple stages where you're validating changes before it actually goes live. But I think often when I talk to people about this, they think there's some kind of inherent like either or of either you can use AI or you can't. And what I found is even in those mission critical production workflows, AI can still help me generate a bunch of unit tests, I'm still going to validate and make sure the tests are right, but it's helping me just move quicker on what I already know that I need to do.

36:46Now, the biggest objection of AI generated tools in general that I do strongly agree with, I don't think anybody has really figured out, is what do you do about the newer, I don't even know if I would call them developers, but the people who are just getting into learning how to output code. How do you help them get over the compiler errors, the stack traces, the debugging. Like, I don't know if that's what you were thinking in terms of the 80 to 90%, but like the rest of the job, I think the tooling there and the ecosystem is still super early. I have confidence it will get better. We will get tools that make that easier.

37:26It's just not, we're not there yet. Yeah, it's like solving those harder problems are almost more like dealing with AI as in a consultant level. And you're posing a high level thesis about what you've built and about what it could do. And then really taking a deeper look at, are there changes we need to make to account for that? Is this infrastructure really going to accommodate it? It's like you're asking a different level of questions. And like you said, that tooling is still early. People are still figuring it out. But what's important is that zero to one is getting lightning fast as you're calling out, which is opening a lot of doors.

37:58And another topic I just want to touch on, because I know this is really top of mind for you, is about opening roles right now and hiring engineers and actually building out engineering teams, these product engineers, as you put it earlier in our conversation. And something you called out to me in our chat is that you recently opened a role that got like over a thousand applicants. Like you're flooded with people who wanted to work at Vercel, which makes sense. Vercel is a global name in terms of its presence. And I wanted to kind of like dig into that a little bit, learn about your experience.

38:28What are like some early ways you're re-approaching the hiring process in like an AI era? So my biggest AI hot take is that, and it spans across multiple different categories, the best practices from pre-AI still matter. And they're actually more important than ever in a post-AI world. So let me explain. With hiring before AI, it was really like referrals were your best source of candidates. They always were. that was your highest quality pipeline versus just inbound applicants. That's still true. It's actually more true now than ever before. And hiring is one category where it works, but that also applies across so many different things that I've just been reusing that for, I don't know, 15 different categories.

39:20But yeah, let's talk about hiring because it is very interesting how the landscape has changed a lot. Simultaneously, a couple of things have happened. One, AI tools make it super easy to apply to just thousands of jobs. Like, of course, before you could go through and click apply and, you know, connect your LinkedIn and submit a job application. And that was great. But you were still limited by human throughput. AI automation, browser automation, using an IDE to like create a little script that does it like much more accessible now, which means that you do see people, especially for technical roles, that spam apply to thousands and thousands of roles.

39:56Which on one hand, on like a human level, I can empathize with just wanting a job and wanting to do whatever it takes to get that job. I think my advice to people if they were in that situation is I think you have the right feeling, but that's not the right solution to that feeling. The spray and pray approach. yeah instead what i would recommend is for those people what i've found and the type of cold emails or cold outbound to me that i get that i i feel more interested in i'm more likely to respond to that i've seen work for other people is very personalized concise and interesting content.

40:45And you can't really use AI to send that. You can maybe use AI to help you refine it and edit it and give you ideas. But ultimately, what I want to know is that this person really cares. They took time to craft something intentional and thoughtful and ideally have brought something to the table that has a hook. It's got my interest. Wow. Look at this thing that this person built. and they didn't try to send a 25-page memo, they said, hey, I built this thing. You might find it very interesting. If you wanna hire people who build things like this, sure, there is no shortage of hiring to be done for great builders.

41:29If you can build great products, you're gonna have a great time in getting employed at a bunch of different places, especially if you sweat the details and you really care about making something that's not just good, but great, and not just great, but exceptional. So that would be my advice for that. The other thing is that I often see, even if maybe you don't apply to 1000 jobs with AI, maybe you're only applying to 10, but you're using AI to like generate your cover letter or generate and update your resume for a hiring manager who's looking at 10s or hundreds or 1000s of resumes, who's also using tools that are helping them kind of refine and focus in on the highest quality ones it is alarmingly obvious if you've used ai um like if you just take the output and you don't really change it there's an increasing number of signals that i'm sure will only go up that say okay this is probably ai generated it started with delve you know if chat gpt would output the word delve every five sentences.

42:35Apparently now it's M dashes. So it was always M dashes. So that that one I'm a little sad about because I kind of like them. And now people think that I'm an AI when I use them. Have to kind of be aware of the zeitgeist there and what's changing. There's more words to what you really want is, again, it's a consultant. You can use AI to say critique my resume, critique my cover letter, critique my outbound email. What could I do better? Could I make it more concise. Is there any confusing spelling or grammar, especially if you don't have English as a first language, if it's English as a second language, you can use that to help refine and curate.

43:13But if you just take the raw output from an AM model and send it over, your application will probably get passed. Yeah, I think that's the key takeaway here. It's such a useful tool for kind of like almost like rubber duck in your resume or whatever you're making, like you could tell it to take the role of the person that you're going to send it to and ask the kind of questions that they might ask you. And it kind of helps you refine. It's useful to use a tool like that and what you've called out almost like a private loop in order to iterate and get that really good version of yourself that you can share.

43:45I like how you called out too that the things that mattered most in hiring, they matter even more now. I completely agree. When I get a message, I can tell when it's AI generated versus when it's not. I can tell when it's a short, sweet personal message from someone something something that i like to do is even if i'm going to use an llm to help me like craft a message or prepare something i love to go in and then like wreck it as a human like let's put some lowercase letters at the front of some of these sentences why does it i need to be capitalized uh sign it off with a funny thing and drop a few emojis in it that the llm wouldn't use not an emoji that corresponds with a word it loves but like one of those obscure yeah my take here is if the best use of ai is if i actually thought that it was written by a human but it actually was ai so if you have figured out your ai curation loop and your edit loop where let's say 80 of the content was ai generated but you still brought your own taste and curation and structure that if you want to use a for that, that's great.

44:47But from my perspective, it still looks like a human craft at it. It's very concise. You know, the wording is super intentional. You've sweated every single word in the sentence. That's the ideal case for me. Like I still use AI to help with my writing and it's great at helping pick things apart. And I love, and I continue to take advantage of AI for that. It's just, you're not done just when you have the first output. Yes, exactly. I think that's the key thing. You have to keep iterating. You have to make it yours. It's a tool, right? You can't send it right off the tool. You need to put your personal stamp on it.

45:21And when you talk about sweating the details, I think that's really important. What are some key things that make a candidate stand out? Once they get past this, they're through the door, they've got your attention. What then about their skill set nowadays really stands out to you? Yeah, so for my role specifically, I'm focused on our developer experience teams at Bercel. And it's kind of more of a niched role than a general generalist engineer. So I can talk about what's worked well for my roles. For developer experience, there's really three parts that I focus on growing our developer community, teaching all the developers in the community through education, and then documenting how our systems work, whether that's API reference or just tutorials.

46:06The type of engineers who work on those type of teams generally have some public track record of their work. And if they don't, it's usually a red flag. So my advice to folks looking to get into this type of work is you should work in public. You should share what you're doing online, whether that's code you're writing, whether that's articles that you've crafted, videos you've made, ideas that you have, V0 generations and apps that you've shipped. All of that is great. What it does is when I get a resume and I'm like, this person seems interesting, I will go Google their name or I will click on their portfolio link that they include in their resume.

46:49In the ideal case, I go out to their portfolio. I learn a bit more about them. I read some of the things that they've added to their blog or some of their posts. Maybe they have a newsletter and I read that. They have a link to the code or the apps that they've built. I'll go look at the code or I'll go try the apps. If I try out one of the apps, like a side project, and it just feels really good. It's so super high quality. The design is nice. They've added these little details that nobody would do unless they really cared about making something special. I get interested. I get excited. I'm like, okay, this person really cares about what they're doing.

47:24They care about the craft. They care about, you know, making something that stands out and they're bringing the world along for that journey. They're making a name for themselves on the internet archive. That is a much higher signal for me than a specific employer or a specific bullet point on a resume. And honestly, like I've, I've seen good and bad from the biggest tech codes to the smallest startups and everywhere in between. So it's hard to index on a specific employer name. Generally, at least for myself, I'm biased more towards like small to medium sized companies. I just find that there's nothing necessarily wrong with working at like the major tech companies.

48:11It's just often I think when you go there, you get indoctrinated into this way of working that is sometimes hard to eject out of. So when I interview folks who spent a decent chunk of time at the Googles or metas of the world, I'm trying to understand, are you excited about and interested in going into the world of you have a 10th of the team size and you don't have that dedicated internal tool that's very, very powerful that a team punch with engineers built? You don't have that anymore. Are you still excited about doing that, But especially more on the like content side, it's like there isn't a video editor.

48:48You know, there isn't another person coming to do this work for you. It's you. You're kind of doing it. And are you excited by that? Or is that kind of scary? So the journey that you've described, I really relate to it because it's kind of like matches my own journey of coming into tech. I had a totally non-traditional journey. And I started by building in the open and building an open source communities, just kind of hacking around with things and sharing it. And over time, you build a collection of stuff that's very personalized to you. And even maybe it's not, none of it's complete. And none of it's even really all that great.

49:19But all of it showcases over time your ability to build skills, to be fascinated by a problem, try to learn it. I still remember in the first interview that I managed to make all the way through to the CEO round. And when I was like really interviewing to get into tech, I remember going into the call and the CEO had opened up one of the games I had made, like in Phaser or something, and was playing it. And was like asking me questions about it. And that was his interview. And I got that job. And so I really can't stress enough the importance of building in the open about just building vulnerably to don't feel like you have to hold on to it until it's like perfectly mastered and ready to share because that I promise you as a developer now that will never happen.

49:56You need to share it. And that's how you can really showcase those skills. So everything that you've kind of highlighted about what makes them stand out, that really resonates with me. Really great call out. And so to wrap things up, I wanted to ask too about for people already on your team or already kind of using these tools, what are some ways that Vercel is thinking about upskilling and coaching people that are already within an organization to succeed in an AI-driven world? You referenced those engineers that are used to being in the big orgs and having a whole bunch of tooling. The same thing can be true within any org that's been around for a while.

50:29People get used to the way of things. So how do you shake things up in Vercel and get people excited about learning? Yeah. One of the things we talk a lot about internally at Vercel and sometimes externally is this motto of you can just ship things. And really what we mean by it is you have the agency to control your destiny. And that doesn't mean that you have to be a developer. It can be anyone at the company. You can just ship that blog post about how our system works. You can just ship that prototype of an idea for a new product we want to build. you can just ship this explainer video talking about how one of our framework APIs works.

51:06You have the agency and the control to actually go and do these things. You just have to have the intrinsic motivation to get up and go do it. So a lot of what I think adapting for a world with extremely powerful AI robots who can just have all of this knowledge of the world easily accessible is recognizing that what separates you apart from everyone else is your commitment and motivation and just relentless pursuit of excellence, showing up every day, getting 1 % better, reflecting on what was good, what was not good, being very truth-seeking and feedback to understand, okay, what did I actually not do very well here?

51:51And please be honest with me. And AI is great for that too. It will happily roast you and give you feedback. And then taking that and like having this, you know, positive, constructive, optimistic view of getting better and going into the future and trying to just be a little bit better version of yourself than you were the day before. The way I like to think about it is not necessarily a competition, even though competition is sometimes helpful for people. For me, it's mostly self and inward looking where my goal is that the next piece of content that I put out, whether it's a blog, code, tutorial, video, whatever, the next thing that I put out should be my best work ever.

52:34And the one after that should be the best one I've ever made. And after that, the best one I've ever made. It's an audacious goal. It doesn't always happen. Sometimes I miss the mark and I don't do as well. Or sometimes I just accept there was a time crunch and I want to get this thing out. And, you know, I'm happy with the quality given the time constraints. But that goal, that vision is still what I want. I want to continue to produce stuff that I'm proud of. As if nobody was looking, as if I was just doing it for myself, would I rewatch and be like, damn, I am really happy with that. And I think if you apply that to everything across engineers, product managers, designers, product marketers, anyone else who works on technical products, you can build great teams who are really motivated in building things they care about.

53:20It starts with caring, I think. Love it. I love your obsession with building and for going for the details and for iterating in the open. This has been a really eye-opening conversation for me, a lot of fun. There's been so much in here to unpack. Lee, we're going to have to have you back on here in the future to kind of check in on how things are evolving because even now I'm like, I'm going to go heck around in v0 i'm quite interested but uh just to wrap things up i i also wanted to ask you know where can people go to follow you learn more about your work and what's going on at for sell yeah uh for all things lee uh lee rob.com it's got links to all of my stuff and for sell.com if you want to learn more about for sell v0.dev if you want to become a vibe coder and just build your next idea um yeah amazing well we'll put those in the show notes for our listeners and And hey, if you've been listening and you made it this far, then clearly you loved this episode.

54:10I did, too. It was a lot of fun. So be sure to check out the conversation on Substack with our newsletter if you're not already reading it. We publish this every Tuesday. And Lee and I are also going to continue the conversation online. We're both on LinkedIn. So please ping us, reach out to us. We'd love to hear what you thought about today's chat. Until next time, that's it for this week's Dev Interrupted. See ya.

54:36Bye.