Vercel’s Big AI Bet. ft. Malte Ubl | S02 E09

5 months agoMarch 10, 2025
47:59
1,887 views
48 likes
B

Backend Banter

Channel

Interviewed Person

Malte Ubl

Description

Today we welcome Malte Ubl, CTO of Vercel, to discuss the evolution of frontend development and the role of AI in shaping modern engineering workflows. Malte dives into V0, Vercel's innovative tool for generating frontend code using AI, and explains how it empowers developers—especially backend engineers—to build full-stack applications. We also explore the concept of verticalization in engineering roles, the challenges of balancing innovation with infrastructure, and how AI is redefining productivity in tech. Learn back-end development - https://www.boot.dev Listen on your favorite podcast player: https://www.backendbanter.fm Malte's X: https://x.com/cramforce Vercel: https://vercel.com/ V0: https://v0.dev/ Timestamps: 00:00 - Intro and Meeting Malte 01:23 - What is V0 and Why Did Vercel Build It? 04:00 - The Technical Challenges Behind V0 10:34 - How V0 Leverages Tailwind and TypeScript 19:15 - AI's Impact on Frontend Development 23:20 - Verticalization of Engineering Roles 32:01 - How AI is Changing Development Workflows 37:09 - The Future of Full-Stack Engineering 44:08 - How Vercel Balances Product Innovation with Core Infrastructure 46:06 - Final Thoughts and Where to Find Malte Like & subscribe for the algo if you enjoyed the video!

Transcript

Group by:

we're in a world where you just use whatever the AI likes yeah I probably need to stop on typescript so much considering we are launching our typescript learning path here well I think today is really obvious that the actually the best Engineers are the best users of [Music] AI today I am joined on time for the well first time in a while it's interesting interviewing people that are coming from businesses they actually show up at the allotted time which is

pretty cool malt welcome to the or Malta I'm sorry welcome to the podcast thanks for having me super excited to be here okay so you are the CTO of versell I'm not getting your title wrong am I that's right yeah I've screwed that up before I don't want to I don't want to mess it up again super excited to have you on we're going to chat about AI we're going to chat about VZ which is I mean has a lot to do with AI but also has something to do we have like an AI counter in the top saying how many times we say AI in 30 minutes in an hour and it's yeah we

might overflow you know a 16bit integer yeah we'll see how it goes but let's just start there so I've used verell in fact literally just this morning um we just made an acquisition of I code this uh it's Floren pops website for learning to code and he was transferring me over his versel account so I was in there messing around with stuff again and the whole product idea behind VZ is very interesting to me because it's not something that I expected from versell it's something that I expected from GitHub right it's something that I

expected from Source graph versell in my mind has always been basically an infrastructure product can you tell me like just starting from zero why you guys decided to ship vzer and I guess what it is for our listeners who don't know totally yeah so vzer is a tool where you put in a prompt and it produce for you initially basically a design with the react code behind it and these days kind of a full stack app

potentially connected to your back end that just implements that prompt and you can be very basic just make me a landing page or you can say here's my open API back and it will go and build you an entire Crut app uh maintaining that that API it really kind of does this like end to- endend EP generation but it's called vzero because it's not meant to do the whole thing that you like live in in prompt land forever but kind of get you started because I in my experience it can often be very daunting to make that first step and so starting with something that's maybe

70% 80% there can be really fun and and and energizing that's that's where it comes from I think the the origin of the products were really that when when CBT came out you realize okay you can prompt these things they're good at text they should be able to make HTML that's not a crazy idea right lots of people were cooking on that so we were just trying it out and and I think we as everyone else realized doesn't really work all that well now nowadays you could do this right but like CH bt35 was just wasn't quite as good like I literally remember like someone in the room like raising their hands like hey oh my God I I figured it out what they figured out is that there's this CSS adjacent

technology called Tailwind which is all in line and it's just old enough to be like in the old jbd knowledge cut off so I knew about it in line like the the llm doesn't have to reason about like I write the CSS no I write the HTML it's all far apart like I forget about uh things so instead it does it all at once and it does an incredible job and so that's how we went from like okay we have nothing to we now have something actually does a decent job and then and we can go from there I'm really glad

that we got to this part of the discussion which is like basically what the technical challenges were because I've ran into this exact same problem where I have had incredible success with llms of honestly like all the models 40 anything from anthropic even the newer reasoning models and the idea of proximity of where the logic lives is like the biggest barrier to me using these tools for anything more effective than just like spitting out a function right and I can see how the of like okay I've got an HTML file and a CSS file they need to cross reference each other

96 segments (grouped from 472 original)10415 words~52 min readGrouped by 30s intervals

Video Details

Duration
47:59
Published
March 10, 2025
Channel
Backend Banter
Language
ENGLISH
Views
1,887
Likes
48

Related Videos