Real-world AI agents with Vercel's eve (our actual setup)

Hey folks, how's it going? Hope you're having a great day. So I want to talk about Eve. And if you don't know what Eve is, it's basically a way of building AI agents. Whether it's kind of like on your website, whether it's doing changes, or it might just be a chatbot within Slack or something. Um, I want to give you kind of like the top level overview. I don't often see people show real world examples of these things. They're very nice in a demo environment, but those demo environments are great until you actually use them in the real world. So this is a real world example. This is where we're using Eve on our own website. And something that you might want to think about if you're doing like comparison pages on your own website. This is a great way of doing it. So we have all these different URLs. Um, I'm hoping we don't get in trouble with Google at some point because these do have like a programmatic nature to them, but we try to keep them up to date and fresh with very organic information. Um, so if we're looking at these pages, we have this last verified date, which is the 17th of July, which luckily is today. Um, so this for example is going to take information called challenges with WordPress and tell you all the challenges that WordPress has and then tell you how good Sanity is because of the URL structure. It's got the problem platform that we're moving away from and the solution platform that we're moving to. Now we generally do a hell of a lot of this kind of work. We do a lot of headless CMS migrations. Um, but we have to try and keep these up to date. Now we use YAML behind the scenes to be able to keep all that information in one particular file. And the beauty of having a file-based system is that you can keep these updated and you can create pull requests to change things inside of these. And so this is a perfect example of something that you want to do with Eve because it's versionable, it's simple, and it's kind of very deterministic. And the reason why I use the word deterministic is because if something is deterministic, it's way easier to do it with AI and it costs way less money on tokens. And we'll come back to that at some point because token spend is going to be a hot topic. It probably already is, but in the next six months it's going to go through the roof. Um, but you can see here this is getting updated. Now the second part of Eve, and this is kind of like an announcement that came along with Eve, which I would argue is probably the most important announcement, is this connect, this vercel.com/connect. Uh, this is a way to be able to make these agents just integrate directly with whatever tools you're using. Uh, in my particular case, it's Slack and Linear we use for the vast majority of time, but you could use this with GitHub. You could use it with any of these. You can see here just below, you can click any of these and they'll work, which is really neat. Uh, and all that changes here is just throwing this in and it's all connected inside of Vercel. Now let me show you in Zed um some of the findings in here. So this is basically like an example finding. So we have uh this updater bot that basically goes out, uses sub-agents to find all this information and create a pull request um to update that information. So in this case, it's like, oh, here's the Framer FAQs. Here's some uh official URLs where we can have a quick look at all the different Framer things. Um, and you can basically set these things up. This is inside of the actual agent skills. So we have these agents. We're a little bit, we're sloppy on the naming convention right now. It's going to get better and we'll release some guides as to how to do this. Um, but essentially right now we have like negative features. So no self-hosting. So Framer sites are locked to Framer hosting. Uh, here's the sources where this is coming from because what you don't want to do uh is put all this information in there that's incorrect without the sources, without it being generated. Uh if you have all these sources in there and you're generating this stuff, I actually personally think it's better than a human researched uh system because it knows exactly what it's looking for. It knows the fors and againsts. Imagine what you can do with this with comparison pages, you know, from one software company to another. Everybody does it. Don't get me wrong, it's not the most ethically like outstanding, really really nice way of doing business, but it is something that pretty much everybody does comparison pages. Um, but yeah, you can see this is an example finding and then you can create skills based on these. Here's like the guard rails. Um, so skeptical voices, all kinds of different things you might want to put, you know, like this is uh the way that we talk, this is our messaging. There's going to be something if you go dig through this. I bet I've showed something here where I shouldn't have showed it, but you get the idea, right? Um, and so what's interesting about this is once you've got all this together, you can create a loop that basically keeps going and keeps going and you can set it on cron jobs. Uh, we set a whole bunch on cron jobs inside of here. Let me just move this aside and find it. Uh, where are we? There is a cron in here somewhere. Schedules. So you can see CMS freshness, for example, is on the cron job. Um, and I can never read these things, but this is on I think once a week and I think it's on a Thursday from memory. Um, but essentially every time it goes and runs through that cron job, you're basically updating across all the different CMSs that information. So you've actually basically created a self-updating bot. And guess what the bot does after it? It tells you this information. And so once you've built one of these things end to end, it then becomes very simple to create the next one, and the next one, and the next one. And you can either do it in two different ways. You can either make them in isolation. So you can have all these different agents that don't really talk to each other or you can do like we've done here where we have, you know, like competitor intelligence. We have daily SEO, we have a Reddit scout, uh we have like route strategies, uh CMS sweeps, all kind of stuff. Like there's a lot of information that's in here. Um and again, a lot of different skills. So I'll show you just a very quick example inside of our Slack. I'm trying my hardest to not accidentally like dox anything or put anything out there, but if you look inside of Slack here, this is a great example of it. So um, inside of here, we actually have a website report bot. And this website report bot takes information from um, Vercel observability and a whole bunch of other different touch points and basically tells us what's going off. And because we've got all this information ahead of us, we can actually make educated uh decisions as to how we progress the website. So this is a really cool feedback loop and it's been pretty much vital for our growth as a business. Um, where you can continuously spin these things up, continuously rely on these. And the best bit of it all is I put like a hundred dollars on our AI uh whatever it is, the AI gateway. Um, and I haven't spent it yet. And we've got like five different um agents that are constantly running. And we've been doing this since pretty much day one of Eve dropping. Um, not Eve dropping, but like Eve the tool dropping. Um, and so it's yeah, it's pretty pretty wild. It's pretty cool. Um, and I would definitely recommend that this is going to be the way that people use agents more with being able to take all these different disparate systems, pile them into a single Slack channel and not making it overwhelming. Um there is a different way of doing this completely separately as well which is kind of interesting which is you can use things like Claude Schedule but the problem with this is that Claude Schedule and I can talk about this in another uh video if anybody's interested but Claude Schedule is a non-deterministic way of doing this and it's pretty difficult to get it to be deterministic. So it means that you get kind of like instead of rigid things where it tells you and being able to kind of piecemeal get better and better. It's kind of like a scattershot where it might answer this way one week and it might answer this way another week. And the problem is it's very difficult then to you know take that information week by week and compare it to one another if it doesn't match the same data structure. So I'm pretty heavily, I'm I hate this expression, but I'm pretty bullish on Eve being like a long-term uh tool that we use internally to take all these different systems, mesh them together, and just tell me what the hell's going off on the website rather than being completely clued out and not understanding how, you know, like Google Analytics works, Google Search Console works, uh Post Hog, Ahrefs or whatever, different disparate systems. You get the idea. Anyway, if you want more info about this, drop us a comment down below and we'll uh we'll get back to you. Thanks for watching. Have a great day. Take care. Bye-bye.