Vercel Ship 2025: The no-nonsense approach to AI agent development (Malte Ubl)

[Music] [Applause] how's it going everyone we're going super back to back I wasn't prepared to come here already great so yeah welcome my name is Malte I'm the CTO of Vercel and you've probably all seen the keynote today this morning or if you caught up on YouTube probably has been in your in your day there and and my goal here is to make some of the things super concrete so we talked about AI about agents and I want to say I'm the anti-hype guy in a very specific way

because I'm actually kind of a hype guy but I do hate this one phase in the kind of hype cycle of soft technology where we kind of all have a little bit of FOMO we know there's this new cool stuff i want to build it but we don't know what it is and so here we are and I want to go into agents and give you a way how to

think about building agents for your own organizations so just to get lay of the land how many here in the room have built an agent before okay I would summarize as 3-5% of people so my thesis is that most of you will raise your hand in a year because it's such a transformative technology

because it's a new technology literally every talk about agents spends like five minutes about defining what they are and even better I have three definitions let's start with Anthropic this one's super useful agents are LLMs autonomously using tools in a loop I love this definition the other day actually G our CEO pinged me and was like I love this definition great the thing about it is it tells us what they do

like they how they do it but not what they do right like great great technology but why would I care google to the rescue okay so Google says agents are AI systems designed to perceive their environment cool make decisions useful and take autonomous actions to cheat achieve user-defined goals again seems really useful right making decisions

taking actions so agents of software that make decisions and do stuff great I will contribute another definition actually it's not really definition but I think it's it's helpful to say from my point of view agents are a new kind of software that we always wanted to write but couldn't for economic reasons right they're a new kind of software and that's how I structure it so my my

in my van diagram which is like kind of stylized here but I have this like really big circle right there's all this software that we would like to write and there's so much software we would like to write right and then there's software that used to be economically viable to write right and it doesn't fill out all of the software that we want to write and so now agents come to the scene and they're adding to that circle there's more software that we can write and it's

very valuable software this software kind of goes through the world it explores its problem space it makes judgment call of how to proceed how to move forward and then can take action so this is very abstractly speaking and we were able to write software like that but it's exactly the type of software that would lead to major spaghetti code right you kind of have this like notion of if this then that except if all these like