Speaker 2 (00:05.527) right.

Speaker 2 (00:10.574) All right. Welcome back to the big cheese AI podcast. I'm your host today. My name is Sean Heiss and I'm here with Jacob. What's up dudes? And we got Brandon. Hey, I, I, before I cut the music out, I just want to say I missed you guys. It's been what?

sub.

Hello everybody.

and she's been.

Speaker 1 (00:32.731) Two weeks? So we're kind of doing the every every two week podcast right now for the summer.

Two weeks.

Speaker 2 (00:40.34) Well, we're ready to go. So so today, are we? don't even know. Well, we're today. We're you know, today we we're on a long journey and we're on and we're all continuously learning. Next week, I'm going to be on the same stage as the governor talking about AI. Really? Yeah. It's no, it's for it's like a thing for economic development, but it's run by Morgan Stanley. But.

rally thing?

Speaker 2 (01:08.652) That's not what we're here to talk about today, even though it influences every single topic we say and discuss, which is, you know, we, what vibe coding has Brandon cracks a drink and tries to make. I'd coding is what we're talking about today. Okay. Not vibe coding, vibe coding, not vibe coding our journey. So there was a bunch of hype, right?

I was trying to be soft. Not vibe coding.

Speaker 2 (01:35.084) We went and we did the AI summit and we talked about vibe coding. You did the 10 commandments of vibe coding. number one, and I think the definition is changing for us. So we're doing this now in production with like real world examples. And so I think we have to refine the vibe coding discussion.

Number 11 is don't do it. Just kidding.

Speaker 2 (01:58.146) from a what you can do maybe with lovable and prototyping versus like actually trying to build something that is going to scale into the future. So Brandon throughout the idea, I forget what the slack post was, but it's like.

why we're not vibe coding anymore than something like live long vibe coding.

Right. It's like vibe coding sucks. We're all about it. Right. And I want to kick it over to you because I think you've got a you know, is your idea. You've got a good perspective and then we can kind of react. But like from where you started vibe coding to where you are now, like how is your perspective changed on the concept and how are you thinking about it today?

still.

Speaker 1 (02:38.766) So when it first started, I completely believed that if you were a developer, you would be able to build some really crazy stuff, like really powerful stuff, very quickly. That's still very true, but it takes you to 80%. And it's at 80%, you can get from zero to 80 % in a day, damn near. It's that 80 to 100 % that's friggin' nearly impossible. That's where it really starts to break down.

me, the whole concept of VydeCoding purely as where you're just letting the AI do its thing and you're just, you're not coding at all, I think is a recipe for disaster at this point.

The and now maybe as the models continue to get better and better and better that's going to that's going to become but right now we're just not there like when you have these huge code bases in your not like there every single action that you do has to be very specifically defined and laid out and if you're not doing that that things going to go fucking crazy on your code base you're going to have duplicate code all over the place you're going to have a bunch of wild stuff happening that you never even I just had it today I was like a go I struggling with this feature I'm like fuck it I'm just going to let a I build this one feature and so finally after

yelling at it a lot.

They got it kind of working and then I go and I start looking through my code and it created files all over the place. It wasn't like it didn't follow anything. I'm just like what the hell is going on here. So the fact that that you just you cannot trust it. The other piece that really upset me is as I've talked about a lot is that I get very angry and I get very vocal when I'm interfacing with AI and I'm very mean to AI. If you're really mean it eventually is just going to be like OK I'm just going to go I'm going to break everything just to get this

Speaker 1 (04:21.712) this one thing fixed so you can see that we're just gonna get the bare implementation of it. And then it goes and it just removes all of your database security, all of your row level security, all of those things. It just removes it because it just wants to make you happy. And that was my marker where I'm just like, I'm done with this. So I started, I probably tried to build.

15 different applications since we've really started kind of talking about by coding cursor and I got into cursor all of them I got to about 85 % and then most of them I just could not finish because like they were in languages that I didn't know right like roster Swift or whatever and so I was just like yeah I'm done with it so the idea of 100 % by coding I think is a fool's errand. It's going to get it's great for prototyping like we were saying it's great for coming up with ideas it's great for coming up with like how you want your user experience to be but then when it's time to actually implement this stuff.

You're just going to have to go back to traditional development best practices and and do not trust the AI

I think the important thing you said there is let's not throw the baby out with the bath water. Like just because we're not in a place where you can't click a button or give it a one liner to do an entire app, which obviously is ridiculous, but you know, it was doing some pretty impressive things with just a few lines of text of like, here's roughly what I want to build.

The opportunity is in the non developers and the designers and the people who are requirement gathering and like ideating. like the concept of like in the beginning of a project or the beginning of an idea, like you want to get as many iterations as possible so that when you get to the developer, you can make their job a lot easier or like we keep talking about, get further faster or get to a better end product. Right. So that I think the important thing is will business.

Speaker 3 (06:12.01) owners and product owners like realize that just because we can get a quick prototype that doesn't necessarily mean matter of fact, it may make building the thing take longer, but you're going to get to a better end point. So I'm not sure anyone or people will realize that, but that's what I agree with too.

So I think product managers are probably going to be some of the best suited for like lovable. So like what I've been doing is we'll have like a UI for one of my clients that I'm not working with anymore. They would be like, hey, we need to come up with a new pattern for this or this or this. Then it was like, just go to lovable. Let it and just be like, just working on this one project's literally just a component. I'm just having you design a component. And it goes through and it actually does a really good job. And then you can be like, now that I see it, I can play around with it be like,

That's not right. I want you to tweak here I want you to tweak this and then at the end of it you can actually give it to a developer and if it's next and Shad Cn they can go and like take a lot of that and reuse it so I think that that's a very powerful thing that product managers could be using today that for lovable, but you're not gonna build the whole app just build little things to share with the team

Yeah. And what you said about, now that I see it, that we see that all the time when we work with clients is people spend hours and hours designing and thinking through requirements and figuring out what it is they want to build. And then when they see it in the wild, they're like, I actually wanted to do this. Right. And we are pulling our hair out because I mean, we actually, we get it. Like this is what we deal with all the time, but it would be a

It would be a lot better if they got to that point faster to where when they're talking to us. 100 % yeah yeah thank you for meeting me.

Speaker 2 (07:51.678) I made a dumb ass over there.

trying to get my mic set up and it just picks up every thundering boom. but yeah, so exactly.

I think the ability for people to visualize the ideas that they have is probably one of the best features right now. Because again, like we've all said, even when you have a Figma design, right? Like when you have a Figma design, you have this kind of static pixel-based thing, then all of a sudden you see it in mobile or you see it on desktop and you see kind of how you click around with it, you're like, yeah, that just isn't, even a Figma prototype, right? Like it just does not translate very well. So I think we eventually start getting

a place where potentially like and figmas kind of going down that path maybe

Yeah, they have their whole like they can build websites there, right? Right.

Speaker 1 (08:39.329) But yes, and I've not played, have you guys played with that at all? No, either of I, I haven't even looked at it.

I think with the my my just listening what you guys are talking about with vibe coding my my take on it in general is that there is a big chasm of expectations where you go from the 100 % vibe coding world to the reality of code and production. And I actually think if you want to completely construct something and vibe it and really just iterate on what the results are giving you from

not necessarily from a code perspective, but from what you're actually experiencing and seeing as, you know, as a user of what it's outputting. I think that what happens is every single prompt, you get further and further away from being able to create maintainable code because it's just going to keep churning. Like you might have, you might have multiple sections of the same type of thing that game implemented slightly differently. Right. And so you really have to have a human in the loop.

this entire time, if you're talking about doing something in production. Because the people that don't code have no other choice but besides 100 % vibe coded solution. And even as a professional that has written code for years and years and years, you see that what's coming out of what I'm prompting isn't very good. just think about what's something that has no idea what they're doing is getting.

Now I will say that, I've been, Bob Maddox is a friend of mine, he shared, yeah my light went off, he shared a starter prompt that he's been using that was actually really amazing. I mean it went through and it broke down like all these rules, all these regulations, all the things that it should do and it shouldn't do and blah, blah, Not sure, I will say that the projects that he has been doing feel very complete.

Speaker 1 (10:34.128) security side I don't know I'm not looking at the code but he's actually been written and it's very possible that you're going to find some people who who are going to be able to do this and be able to do it really well better than me you know because I just get angry I get frustrated I want to flip my desk over right that there might be people who who end up being able to do 100 % vibe coded I mean he built it in like eight hours or whatever it was and it was a pretty cool like it was a pretty cool application I mean it looked pretty good and all that stuff and so I there might be those people that can do it I just

I haven't seen it yet and everybody I talk to is the same kind of thing where you just get to that 80, 90 percent and it's just like the wheels start coming.

So I had a project that we took on that was, know, or basically required a decent amount of resources in a short amount of time, but it was kind of right in my wheelhouse. And one of our developers had to be on a different project, and so we had to get it done quick. Sorry, my microphone is absolutely.

We're having technical difficulties today.

I guess long story short is we had to build this thing super quick, but it was something that was going to take the quote for the client was like this is going to take five, six weeks. Okay, and it is one of those things where it is a very large multi dimensional thing. It had it had real time syncing from a database rehydration of a really complex form that had lots of external API calls and all this stuff.

Speaker 2 (12:07.606) Okay, got to that 80 % on Friday. The plan was to turn it over on Monday. Well, Monday got into it and realized that last 20 % was not going to be able to be achieved. Okay. So, got in there and basically had to go back to the very fundamentals and beginning and say, okay, how, how was this built? How can I, how, as a human, how can I break this up? And, and basically grinded

without stopping for like 10 hours. But then I got over the plateau, got over the hill, right? In my opinion, I never really thought of vibe coding as 100 % vibe coding solution either way. But I feel like a lot of people can get anything to 80%, whether it's vibe coding, whatever you're doing. I think that last, getting to 80, 90%, that grind, that discipline that it takes to stay in it.

until you're complete, because people love to build little ideas and then when they kind of get it done, get it going, they kind of get bored. But if it's the people that have that discipline of that last 10, 15%, and then they get over the hump, and then guess what happens if you have good structure?

It's still gonna fucking measure CODA!

It's it's it maybe is but it maybe isn't it will hit it. But here's I think it's I think you can totally mitigate it. I think you can totally mitigate it. And one of the problems I'm having.

Speaker 1 (13:28.494) I promise. Okay.

Speaker 1 (13:38.478) No, you can't. Yes, you can. Let me tell you why. Real quick.

Yes, you can.

So I realized that OK AI is really good at doing various tasks. I can give it an example and say hey go do this because I don't want to do this five times right. I've got a legal document a legal markdown I got a privacy markdown I've got a terms markdown right. I've got three different pages I need three server side little calls to basically load it up and to return it blah blah blah. So I do one of them manually and then I go and I drop the folder thanks to your little tip there I drop the folder into cursor and I say hey go look at this and now just go repeat for privacy and terms.

very fucking basic, right? The structure's already there, it's already good. Guess what it does? It goes in and rewrites it for spelt for, not spelt for.

I see what his problem is. He's using a non-widely adopted free-

Speaker 1 (14:25.998) No, so Svelte is your number three framework that's out there, right? And I mean, the fact of the matter is, is Next, the same exact thing is going to happen with Next 17, right? What's going to happen is these large language models are not going and people are like, well, just include the documentation. Good luck.

No, it won't work. I agree with that. When so one of the one of the problems with by coding is his latest and greatest stuff. Dude, the project that I'm working on right now is the back end is written on Ruby and it rips Ruby like it just absolutely shreds. Dude, I haven't had to touch it like haven't touched it.

Like in a good way.

Well has there been a Ruby update in a while?

I have no idea. mean, we're on, you know, whatever version, but like it was able to. There's certain and literally I will pick stuff based off of my experience prompting it was like like an example. So for for forms react took form, it knows react took form. I don't know why, but it fucking does. And that thing will rip forms. I haven't touched the form code, right?

Speaker 1 (15:31.054) But again, do you want your AI directing what technologies you use? You're fucking crazy. No, because then every fucking design is going to be stuck in 2003 because that's what all LLMs basically come out of.

I don't. I do.

No, because

Speaker 2 (15:47.63) talking about you're talking about svelte no your user is no idea if your app was built in svelte or react it's a fucking web app you can make it look however you

But again, if you actually care about developer experience, you don't use react. mean, that's yeah, absolutely. Yeah, reacts dog shit in terms of developer.

Well, what if

That's not a real...

It's it's really it's it's it's really up to it's really up to the developer, right?

Speaker 1 (16:11.85) No, exactly. Exactly. It's up to the developer, not to the fucking AI because they don't know how to go and update. So you blame Cursor for not doing it. Like Cursor has like some of the rules things. Cursor rules as we I've talked about, it's a fucking joke, right? Like they almost never work anyway.

But the bottom line is that there's your you're you're being washed into the the the riptide of what the reality of the future is which is a people are gonna have to use what what AI

I know. You not see what that what that does right like that that sticks. be stuck in 2020.

It means that that's

I don't know if it translates to the UX. It translates maybe to the developer.

Speaker 1 (16:50.478) How is is UX and how is how is UI development any different than developer developer? Go to developer development. How is it any different?

I mean, those frameworks don't have fundamentally different capabilities. They have fundamentally different developer experiences.

Sure, but again, UX being the same thing, right? Like if all of a sudden we're just letting AI basically do all of our UX design, and it's just gonna continue to do this, anytime that, so I'm gonna go build.

I'm not talking about UI, new UX design. That's a totally different topic.

not really, because again, the only way you get to UX design is through code. so where does the code come from? The code comes from the LLMs.

Speaker 2 (17:29.676) You're micro analyzing front end JavaScript frameworks in that context. Like, like that's that's not going to be that much of a difference. Like, like, does it does it know how to implement like

No, no, no, no, I'm not saying like it doesn't know how to implement it. What I'm saying is that from a design standard aesthetic, which is all done in code, and that's what these large language models are trained on, is going to get stuck. But the same thing very well could happen with any.

I agree with that. Right. I agree.

Speaker 2 (18:02.764) That in a world where you're not see I come from a world where we we have to match the design from a UX designer like they are creating a mock-up and it has to look like that right like I can't just rely on the AI tool to like be like hey nice Shad Sien looking fucking well

You know, but let me ask. So once you have all the designs done and you go and you start building things out, how often is like, we have this new feature, right? How often are you going back to Figma to do the design work versus just kind of doing it within the environment?

We always go back to design and put the pieces of the puzzle back together.

Yes, and I could see that that's going to change.

it is right. No, and honestly, it's it already is changing. OK, it is changing. Doing it more on the fly, but I still have mixed experiences with that because.

Speaker 3 (18:54.882) Yeah, so, okay, we're talking about different phases of a project too, right? Sure. Like if you are in your experimental phase, you're, you know, kiss another boy, whatever you want to do. but like, just a college guy. Yeah, just college. You can play around. Design doesn't matter as much, but the frustration from his part is like he's trying to experiment and get new and unique things. And you're right that it's not good at that. It's good at giving you the median, the middle point of whatever

modal or whatever. What you're talking about is, wow, for your use case when you were trying to implement some pretty complicated back-end logic that's been done a thousand times, so it's pretty standard how you're gonna implement a database schema to accomplish X, Y, and Z, which what you're doing, really really good at that. You're doing new frontier shit that you're not trying to have a starting point of a figment design or anything to base it off of, which it does appear to be pretty dog shit at right now.

Well, but even if you, so like let's say that I'm not wanting to be bleeding edge and I just want to go and just start a new project. I'm not even a developer, right? I'm just going to go start a new project. Guess what it's going to do when it says npm install? It's going to install the latest. It will always install the latest. Like unless you know to go back and say, only install Svelte kit one and Svelte five, do only do that. It's always going to default to the latest thing that's been released into the npm library. So you're going to have to deal

with.

But I've had different experiences too where like for example on this project there was two APIs that it had to leverage that are external. I fed it the docs from those APIs. Now those are REST APIs different. know what I mean? But it one shot them.

Speaker 1 (20:37.176) Well, but because like rest like the technology of rest, that's just like Ruby, right? That has been done a billion times. didn't know. Feed it real quick. What did you did you feed it? The did you feed it like a swagger doc or did?

And I guess the point is...

Speaker 2 (20:51.234) So here, what I typically do is I go to chat GPT. I have a conversation with it about what I'm trying to do. I feed it, you know, URLs, documents, whatever. And then I get a cursor rules file, which, you know, ends up having, you know, we can have our own conversation about that, but it'll it'll end up having things like example responses, what, what endpoint to hit. It'll be a little bit more like, like built out. Right. But that's if I'm doing something that's like,

I'm starting the new feature. So I come into the new feature with a, the output of a chat GPT conversation. And that gives it a little bit more direction from the beginning, but still it's going to veer off. like you said, you said you wanted to rinse and repeat, right? Okay. I rinsed and repeated the same thing six times last week. It did it six different times. Yeah. Six different ways. Six different ways. But

Okay.

Speaker 1 (21:45.528) Six.

Eventually, it's like, here's the grinding aspect. And I think this is what's hard. And this is what's really frustrating is that in cursors in a specific one, because sometimes it'll just drop your context and you just feel like you started over. But like, like if you find that it's in the it's in the slot, you have to stay on there and keep going. Because you know what I mean? Because sometimes it hits a hot streak.

Yes. Yeah. Okay. It just hit every single fucking.

Single fine. So, so, so I mean, I'm serious. So like, was struggling, but I was like, I figured out, okay, here's the template. Yep. I was like, I had a list. I had initial state. I had the edit state and I had it like really dialed in. Like everything was componentized. I refactored everything that the ad written and I was like, rinse and repeat this and it hit it. And I was like, oh shit, it did it. And then I was like, okay, but, but, but fix the two bugs. And then I had to do it again. And I was like, I think it got to, so there was eight steps. I think it got to the sixth step.

And then at the seventh and eighth, it totally fucking freaked out and lost its mind.

Speaker 1 (22:43.726) I wonder if it is context size that's actually the problem. because like you have no mental model of what's actually being sent to the AI when you're using cursor, right? Like you have no idea. Like conceptually.

Sometimes you'll mention a file and then the next prompt it's gone.

Yeah, exactly. Like, and so so there's just so much that I just don't understand.

I think that I think that you bring up a good point, which is the the you need to know what's in the context window that's being provided and and it should probably have a little bit of a we should probably like see like this. You know, I don't know how you visualize that. I mean, there's two.

too much. Too much. It's too much for a human. thinking about it. I'm like, okay, every time I chat with them, it's going to show me what how the steps now they're starting to do this already with the thinking models of like, okay, I'm going to go look this up and like showing you the process. But that's what you're talking about. It's like, show me your work basically of like, what do you what are you basing this off of?

Speaker 2 (23:41.696) Well, just know that here's the biggest red flag in cursor when it starts searching, searching, searching. God. Stop. No, no, no, no, no. We were already there. We were already there. We were there.

guys.

Speaker 3 (23:53.166) Where did you go? I? I it did something to me this week. I did that same shit where it was like it was like 30 searches later. What are we?

are we here?

So my strategy, I mean, let's be honest, I'm using, there's more critical mass around the tech than in this story that I was using. So there's a little bit more maybe context or experience of the AI. Maybe it was a little bit more focused, maybe it was a easier even. But like really, really, really, really spending way more time than you probably think you need to on making sure you're mentioning the right files, you're saying the right things. And like, it's gonna take a little bit longer.

but you're probably gonna get a better result. And then once you get to that 80%, you have to fucking keep going. You can't give up because eventually you will get over the hump. And I think that part of this too is that like, I forget who is saying the story is like, like, like the Pacers, the Pacers are in the Eastern Conference Finals, right? They went to the Eastern Conference Finals last year. Well, last year, it was a huge surprise that we went to the Eastern Conference Finals, right?

This year it was a it was it was was was you know, it was so surprising to the Eastern Conference finals, right? Next year, if we don't make to the Eastern Conference finals, the coach is probably going to get fired. Right. Right. Right. Because so with AI and by coding your brand and your expectations and all my expectations are through the fucking roof. Right. Like we have changed the goalpost is moved and and we expect more, expect more, expect more. And there's just there.

Speaker 2 (25:30.092) I mean, so we're probably getting more.

Yeah, right. 100 % you know, so so as I shit on this thing, it's important. It's important to note that like, all of the things that's doing still is absolutely mind blowing. It's I love it. I'm all about it. I do just get so angry. Right?

Yeah, well, the thing so one of our developers today we are we are very much in our at our shop use AI if it helps otherwise, whatever you do you but you should we encourage them to use it can reach out to me he's like, Hey, I did this thing I used it for the documentation of how it works and how to edit it and how to do it in the CMS is like, by the way, AI is is money for documentation. yeah, I'm like, again, keep going back to the well of like, use it for things that that save you time, you know,

we'll find those naturally, also like, maybe we're trying to build a house with a, design a house with a calculator.

We've seen, I think we've seen the potential of vibe coding. You've seen it. You have shipped an app to the App Store that you And so you see it and you know that it's

Speaker 1 (26:32.854) Paper Keeper is in the App Store.

Speaker 1 (26:37.738) Exactly. That's exactly it. So I know it's possible. And so then when it doesn't work, that's what pisses me off. And it's no different than Siri, right? Like I know that Siri has done these things and when it doesn't do it, I'm just like, this is mind bogglingly stupid. Like how can a $3 trillion company can't get this fucking.

right. Okay, so I was shitting on AI the other day and use case and then I used it for that exact case. It's not related, but we're talking about but we were talking about AI I was using Gemini and I was like, who gives a shit if it can make a calendar invite for me, I can just open up my app and do it. Well, I literally pulled out Gemini the other day and was like, hey, make me a calendar invite. I have a card oil change at whatever downtown car care on Friday at nine o'clock. I was like, okay, I go in there. It's got the link to the location, the actual

Jim and I made a Google Calendar event for me with a link time and all that to all the details to this thing and I was like, oh I was on a walk with my dog and I two hands but I I remembered it I was like, oh shit I didn't put that on my calendar yet. I'm like, ooh, ooh I have downtime sometimes when I am thinking about things but I'm not at a computer or I'm not you know able to use both my hands.

Are you interfacing with Gemini? So you have the Gemini app. Are you paying for it? Okay, how much is it?

I am. 20 bucks a month, yeah.

Speaker 2 (27:59.126) I don't want to know about Jacob's monthly.

He watched me pull up Claude. He's like you're Claude boy now? I'm like boy. I got every fucking app

max over here at 120 bucks a month.

Dude, dude, did you know I'm not going to. OK, but did you see the new ultra Gemini Ultra with the VO3? Oh, OK, I want that really bad, but it's 250 a month. So that's that's the video. Yes, Sean said. Yes, it was the it was the fake pet, the fake dog one. It was like the.

Did you do it?

Speaker 1 (28:33.422) So if have it I'm gonna if be if what is it vo3 yeah, if you feel three can can maintain character Static yes continuity. I will be paying the I'll be paying whatever I am because I'm going to I'm going to market basically offering kind of like, you know brand characters as a service and and Vivo or Vio 3 is the one I'm kind of eyeballing it

How is that how is that going?

It's been very fun. Shit tons of learning.

But this is some of the vibe coding stuff you're doing right now that you're frustrated with.

No, so I haven't done, I've actually not been doing a whole lot of vibe coding. Like, well, again, what's the definition of vibe coding? Right, right. anytime, so vibe coding originally when it first came out was that you were just 100 %...

Speaker 2 (29:22.862) I never really, I never really considered it that, but now that I, that's one of my things. It's like, you really embraced that. And I, and I think I thought I was doing that, but I was always more of like,

Higher level, like you're vibing, like you are in flow and you're not stopping to be like, how do I do that thing? You just ask the question and it keeps going.

I always look at it more of like an accelerant of what I was already.

But that's what I'm doing now. With Wicked Forum now, it's pretty much like I go in, I've structured it all, I built it all out, I've got all my own stores, I've done it all. Then I say, hey, go look at the stores. I use the MCP stuff all the time.

That's what I was just about to bring up. So like part of this whole journey is is is the MCPs. Yes. Inside of cursor. Yeah. Which are accelerating. I think we got a couple of good. What are some of the MCPs you're using?

Speaker 1 (30:11.124) So I use the super base MCP the stuff. Yes, stripe MCP is a So I went literally I went in and I'm like, I don't want to log into stripe and do all this bullshit. So just go ahead and

How are you using the Stripe MC?

Speaker 2 (30:24.97) It replaces clicking through the...

Want you to go create I want you to create a base a pro and an exclusive I want year and month and I want no the exclusive ones I need a large medium and a small medium and large and extra large

Yeah.

Speaker 2 (30:43.054) And while you're at it, get me a whole...

It's a...

Speaker 3 (30:50.734) Brain has got a little bit of a...

I've little back issue going on right now and I'm not supposed to laugh. But that has been a game changer. I don't have to log, now I will say that the Stripe MCP doesn't allow metadata to be moved over. So it ended up being like, hey, like I've run this but I'm not.

We should do a podcast episode where Brandon just points out edge cases that platforms don't.

What do you mean metadata?

So like if you're gonna go add a product you want to add maybe the metadata of what type of like what type of plan is it didn't it didn't

Speaker 3 (31:21.411) That should.

That's the problem with MCPs right now is that and this is advice for companies that are building them Don't go build a half SM MCP. No, because you just built an interface to your API. That's probably not complete Yeah, so like if you were in this we're talking about with direct us but or any of these other ones I wasn't saying directors as bad. It's like you don't want to push out an MCP. That's only got a couple features now

Yeah, yeah, I'm late. Are you using the direct as MCL?

I use it today for the first. So here's the use case that I saw that I was excited about. They were talking about like, you can make a document in Google Drive or docs and then you can copy that over into your CMS. Directus is a headless CMS for those that aren't familiar with it. Just like it's like WordPress, it's an admin panel. You can go and manage all your.

It was what he showed me was cool as sh-

Speaker 1 (32:11.232) Amazing. It's an awesome platform.

But I was like, okay, cool. I don't really write a lot of like docs, Google docs, and then have a hard time moving them over into a CMS. Like that's a very, to me, not a huge use case, I'm sure, I'm sure a lot of people had that. The one that I saw though was we build websites with blocks. Blocks are just collections in direct. So it's just a Postgres database that you have a block for a hero, block for whatever, your accordion, whatever. Well, you can use the MCP to say, hey, build me a page.

using my blocks and I wanted to do X Y & Z so I just did a very quick I said make me a landing page right and using indirect ascent said okay so it assumed I wanted to make a page with an example of every single block that we had it did it filled it in with content and publish that shit and I and like it took like about 30 seconds of like I had to click accept of like yes you can read it yes you can create that yellow yeah well yellow do that

there a way to auto run tool in MCP?

Yo, YOLO does. I don't even know how to see. Literally tell it. Yo dude, you're in YOLO mode, go fucking banana.

Speaker 2 (33:12.674) Switch to yellow.

Speaker 3 (33:17.998) Oh my god, but what I was, that was my Kauai Leonard laugh. What I was thinking though is like, oh my gosh, it's kind of tedious to build pages with blocks, right? You can describe what you want and the customer can describe what they want with a design and then you can use chat GBD to infer what that design means, know, like dude.

Take a minute, we're just gonna click.

Speaker 1 (33:42.21) That turns into it. That turns into like what a modern day WordPress admin would be for a client. Yeah. What page do you want? I want this page. want this.

this and it just think of a think of it. I'm on my morning walk. I am a marketer and I'm just going to go and I'm going to say, hey, we need to publish a new page on our website today. And today we're going to talk about, you know, the latest and greatest with, you know, I don't know, right, grapes and tomatoes and making salad. I don't know. And then you just like describe, like have a whole conversation, give it a big old context window. And then it'd be like, cool, we could do a hero with this and I'll

go generate an image with AI for this. And then they get back to their computer and the thing is already fucking published.

Yeah, so that's where it absolutely goes there. And so like all my anger and my rage is purely just because we're at the very beginning and it's not doing the things that we think that it should ultimately do. I think we absolutely get there.

Okay, another complaint that I have that I think probably gives a little insight into where you're at because you're using these tools all day every day. I don't think they're scaling right now. Cursor is, even though I'm in the fast lane now, it's slower than it was. It's not getting faster.

Speaker 1 (34:57.838) It's definitely not.

Do you think it's getting faster no cursor scaling to millions in I mean I don't have millions and billions and Whatever users I don't know how many users I think they went I think you know how that that slide that I can talk about slides getting out of date I think they hit 300 million error. So Hey, I have a call with the co-founder level at 9 a.m. For vibe coding competition, right? Which by the way, I'm like we're doing the vibe coding competition for rally and I'm like

Right. Lovable too, man. Lovable is just hockey.

Nice.

Hell yeah.

Speaker 2 (35:25.164) Let's see what these motherfuckers can do. Right? Because you're in right? Yep. I'm in dude, you're gonna win. No.

I don't win anything. I win life. That's it.

Yeah, you do. But I think it's like, think September 20th. But here's my point is like, I think there's a future where I just don't know if a model that doesn't have more context about you isn't going to win. Like, like, like the fact that cursor dropped my context window.

That's come, when is that, September?

Speaker 2 (35:59.36) of the conversation I'd been having it for two hours. And that didn't get somehow committed to some long term memory. Think about all the shit that I just did that it learned about me. That's gone forever.

There's a few MCP memory models that are showing up now that I'm kind of thinking. Right, where again, you're just going have a dedicated memory MCP. You should be able to connect it to Claude and open AI and how. How is it? is it? Yeah, so I think I think literally it's it's it's just like a very intelligent vector vault of your.

I maybe that's what we need to go down that

Speaker 2 (36:23.502) Or is it leveraging something like that?

Speaker 2 (36:34.19) So somehow it.

Yeah, it's like a localized rag. think I'm probably oversimplifying it, but like it's just a...

Is it just gonna bog down the context window immediately?

Because I think what you do like if you if you'd really architect it properly you're gonna say okay We're gonna have short-term memory. We're gonna have long-term memory. We're gonna have ephemeral memory We're gonna have certain types of different memory types that are gonna come in So when it comes in we have to like first identify what type of memory it is We're gonna store it and then when a request comes in we're gonna see which ones we need to bubble up dude So there's like a whole lot of psychology Aaron shaver was worried has has been worried I don't know if he's how far he's gotten on it But he was because he has like a PhD in psychology, too So he's been working on like the

a memory model. I haven't seen it, but then I've seen some other ones come out that are like, they look really, really,

Speaker 2 (37:19.97) really like that's almost got to become part of that neural network. Right. Like like in how the how it flows through a decision. Right. Like it has to be more sophisticated.

It very well, it should be way more sophisticated than just a rag, right? because rag, know, rags fine. But like a lot of times you might just do a search and you just get like a small amount of results back. maybe your question didn't have enough.

something that's more way more like it's like a ton of a new technology.

Google, honestly, like if there's one company who should build this, it's Google. They've got the scale, they've got the intelligence that is beyond just rag that if Google would be like, hey guys, we're just going to basically release your AI memory model and then you just connect it and then.

And I mean, think, so think about this. Think about a future where. Your AI memory had seen every single line of svelte that you'd written for the past two years.

Speaker 1 (38:12.142) It would know it would know it would know it would know listen Yes, I'm he knows me and it knows I'm using runes. I'm not using this felt writable store anymore You know like all that shit's gone. It's done and You know like yeah, so that that's the piece where I think that the memory model could very well override the lack of a of an LLM knowing all of the new

He would know you. He would know you.

Speaker 2 (38:34.712) Well, because what's it doing right now? It's going outside of you to go get the answers.

Yeah. Well, that's one thing. Okay. So like chat GPT has a memory, right? So when I go and ask it questions, it does sometimes like, yeah, I remember you talked about crafted in your marketing. But Google has one now. Okay. Also, Jim and I has their memory.

Maybe that's what...

Speaker 1 (38:56.818) It's got to be one though that I can connect to all of them. Bring your own memories.

That's what I'm saying. go to then to Claude. Yeah. And I'm like, I find myself having to re-explain. I almost need to either just keep my own. The gold. The gold.

is who's gonna own your AI memory. That's the last...

Dude, and it needs to be an open source. It be an open source.

Okay, but wait a second like Google is really concerned or once your personalization right like they can target us it Why this is such a complicated problem? First of all is like what I said two weeks ago may no longer be applicable, right? So like how do you maintain and clean memories because I can do that. I can be like, yeah, throw that away I don't I don't care about that anymore. Right? They don't fucking know. Yeah, how would they know?

Speaker 1 (39:43.416) But do you need to like so so if you if all of a sudden you we get rid of the limitations of our

You you know, what's funny is like that story where they're like, Oh, Uber, Uber, did you hear Uber came out? They're like, you at a predetermined time at a predetermined place, multiple people meet up to go to a predetermined destination. And somebody retweeted him like, looks like Uber has discovered the bus. But like, is that where we're going to end up getting to? It's like, it looks like Google has discovered.

The human brain. Yeah. It's like you already had this.

Right. no, very, very, yeah, yes, most likely. But someone has to actually build it now, right, and make it scale and make it fast and make it actually work. But if I can pull my memories through any AI model, that would be great. And I think it should be an open source project. I really do. think somebody should, and maybe there already is and I haven't.

Well, there is the open open MCP memory is a real project. Yes.

Speaker 1 (40:42.478) Okay, so that could be because like I think that again, as we know, open source like open source doesn't really ever take off just because like the majority of people can't use.

Well, I I think as much as you say open source, I mean I need to own this fucking exact and not have some proprietary company that sells ads on that

Exactly. And then someone's going to take the open source project and then they're going to commercialize it and they're going to put it into their thing.

Is Linux open source? yeah.

It's uncool.

Speaker 1 (41:09.806) What kind of fucking question is that? Oh, you know, I should have, yeah, that was a horrible statement. Open source absolutely does work. It doesn't get the necessarily the adoption level. You're never gonna have a that goes hyperbolic like lovable or anybody else. And then if you wanna have something that can really impact the rest of humanity, it does need to be able to be distributed.

You said open source doesn't work.

Speaker 2 (41:23.477) Well, you're

Speaker 2 (41:33.454) Typically, typically, yeah, typically an open source project has to get leveraged by a by a engine profit or else it doesn't get critical mass.

Dummies.

So yeah, MCPS.

MCP is open source, right? Yeah. Yeah. So I think what you're maybe there's a protocol for memory. I'm talking. Yeah, no, I hear what you guys are saying, though. It's like, it's hard to get alignment and to make something that's like scales to that size and gets adoption. Unless it were just like MCP like an overwhelmingly like good product or design decision architecture for everybody.

Yep, the protocol.

Speaker 1 (42:14.798) I mean, think of it like even if that provider then they offer you a web interface to your memories, right? Now you can go in and like if I was talking to Claude or if I'm talking to Chad GPT or Cursor or whoever, that it has all-

Apple fun to fumble the bag. We're all using their devices

What was it? Was it Minority Report that the guy is like sitting there listening, watching his memories of like his ex wife that died or something like that? Yep. Is that we're to be just going to be sitting in a room crying one day. Probably next week. I don't need that tool to do that.

hopefully.

Speaker 2 (42:50.08) It is interesting watching my daughter like go through the journey of using like we have the Mac mini at home and both both of the older kids have their own accounts on the on the machine and just like how resourceful they are, especially especially my daughter. She uses chat GPT. She she uses like the core. She doesn't download much. So she uses all the core Mac apps, you know, and she she she just like uses that.

Like that operating system is what she uses. And it's interesting to see someone that young just easily just take technology and just kind of use it for stuff that they're not, she's not using it to do anything besides get stuff done or it's very expressive.

That kid whose dad was the drummer and is at two years old just...

You just know, I mean, we're going to see that same exact experience where we're just going to have these new levels of like skill sets that are going to come out from these young kids who are literally born with the technology in their, you in in in their hands. And so that part I'm super excited for. I'm very I'm very positive about like where this all leads. Yes, there's a funny job posted the thing into the Slack channel. I think it was a New York Times talking about how the Claude CEO is basically saying, hey, everybody needs to prepare for like all

the white collar, know, entry level tech jobs are gone.

Speaker 1 (44:23.31) But that being said, everything that, when we get to this point, everything that the AI can do for us as a civilization has really big impacts, right? Like it can, it can work on balancing the budget, it can work on problems continually. If the government finally would embrace it and just say, okay, we're gonna really go solve this and not just set Elon out there to come up with this Doge bullshit, but actually implement things and work on it, we can get to a point where we might not need as

money to survive and to do those types of things so there's a lot I think that AI is going to basically bring us but yeah there's still going to be some damage.

I mean, I'm scared to death of all the productivity that's gonna get I mean if we increase our productivity over time as there I don't think there's ever been a time that it's been linear to wealth distribution like have we ever increased productivity and in the income inequalities has stayed the same and like you know what I mean like it just doesn't happen like

Yeah, that's a good question. In terms of, I mean, I guess you'd have to kind of stack like how much money people are making versus, you know, the productivity. There's probably some sort of correlation there a little bit, but I don't think I don't think there's a direct line. Like if all of sudden I am super productive that you're going to pay.

What's that what there's certain? The problem with this one is at least like the anthropic CEOs that's a different type of person that gets threatened in this situation than normal. Yeah, like typically in the past it was like, the the person that works in the factory. Now it's like the person that's like working like an administrative and lower level position.

Speaker 1 (46:06.36) Like, because we have been attacking, like we've been we've been decimating people going and becoming electricians and plumbers and like all this stuff. And we all want you to be knowledge workers now. I mean, maybe we do need to pay for that.

I mean, I think that there is productivity shortages in the economy, probably, and there's shifting priorities.

Just answer that question. 1975 was the last time. 1972 was the last time we had any sort of correlation. The chart basically looks like productivity is hockey stick. what are they?

What we're waiting is that wages wages never increased. Right. Yeah.

Yeah, so we're still at what? point seven dollars and 25 cents. Yeah, something like that. Okay, so that's that's fair I mean and and and that is capitalism doing what capitalism

Speaker 3 (46:53.132) That's why I'm scared. I am both an optimist in this regard of like I see all the crazy opportunities of doing more with less and like making better products and like you know having this utopia of like wow we can we can crush it right but then the reality of our current system is it is not

built for this. it's going to eat all profits, it's going to make the rich richer, it's going to make the poor poorer. And that's what our system is set up to do.

I think the biggest threat to the future is brain rot. I don't even think it's AI. I think it's just the rot. It's just the rot of content that gets put on devices that people get addicted to.

Yeah.

Speaker 3 (47:37.176) Yeah, no, that's a good point. You shared that faceless YouTube channel that was like a bear versus a tiger. Who wins? right. It's like, that's fun. Yeah, that is fun. A little bit of that's not bad, but to Sean's point, we can make addictive. We've been able to make things addictive at scale, but not at this.

There's weird things that I think I've noticed like a couple things like and I don't know maybe you guys can share a couple of your experiences but like there's certain niches of content that have been absolutely just Like thrown at me and have changed the way that I act One of them which was pizza I told you guys this a million times last year. I got addicted to pizza. got addicted to making it I wanted to it all the time

and I got really good at making pizza. I mean, really good.

That's a pretty wholesome addiction.

I had to quit Instagram. I quit Instagram for a while. I am back on it now. But I mean, I was just watching people make pizza all day all day long, like at night. I'll just say, OK, you need it this way. OK, need two hours rest. OK, this much hydration like it was.

Speaker 3 (48:49.546) I can't go there. I really want to ask if it got sexual with

No, but like at the end of the day, I got how much time do I need to spend thinking about pizza? How much time do I need to spend thinking about pizza? And now like, like there's micro algorithms, but one algorithm that I've noticed that has had a profound impact on younger people is that I think every single male hetero male between the ages of like 22 and 28.

are all obsessed with golf. And it has 100 % to do with social media and YouTube.

I actually have no problem with that. And I'm almost glad, right? Like if that's really the case, that we have, because the reality is the thing that scares me the most about these younger, the younger men, is that we have this whole social media influencer thing about incels and like, you know, the red pill bullshit and all of this stuff that basically is saying, you know, the reason you can't get women is not you, it's because all women are like this and that they do this and all this kind of bullshit.

that they're crafting like this is one thing that this is one that I land on all the time because I end up getting so upset with it but I still watch it just to kind of figure out what's going on.

Speaker 1 (50:10.53) As we were talking about the moot series I was just watching, A Mindhunter. Where it's talking about where they're starting to break out, what are all of these serial killer men, is primarily what they all are, what their motivations were. And it was their mothers rejecting them, the women rejecting them, all of these things that are very much specific about rejection and not feeling good enough. And then you have the whatever fucking podcast where you have those guys who are just sitting there literally just making fun of women.

This is I've got to say this one because it's fucking hilarious. So the whatever podcast really popular podcast. It's very much a Just a bunch of dudes basically saying yeah women are a bunch of idiots and that you need to have a dominant really yeah, it's it's but it's a very popular podcast. So one of the guys his name's like Andrew He's a complete douchebag but so he had this one girl that they were debating and he gave her an olive a jar of olives to open and she couldn't open it and she's like Yeah, I can't open it you open it and he's she slid it to him and he's

like, you want me to open it? And she's like, well, yeah. And so he couldn't open it. And he's like, well, you got it all greasy. He's like, give me a napkin. You your greasy ass hands were all over it. He couldn't open it. And then he gave it to his assistant. His assistant came up and just popped it right off. And it was just the peak, the most satisfying thing, because it's like, you just got caught. You worthless little prick. But those people worry, those worry me. Because again, more and more men are

Yeah, there's okay. So go back to the golf thing. Like that's a good influence. Yes, it is. And then what's what's what's interesting to me is that like Tiger Woods made a people really love golf in the late nineties, 1997 when he won the Masters, everybody wanted to play golf. But like it's changed to where it's changed to where like people that don't normally see that get to see it.

Right. And I think that's what's what's one of the bad one of the good and bad things about social media is that people started seeing how the other side lived like people that didn't have as much. They actually got a vision into the window of how other people live. Right. And so people grab my issue with it overall isn't golf. Golf's very positive thing to do. I think it's people gravitating towards some sort of specific lifestyle or thing that is like, you know, maybe hard to sustain.

Speaker 2 (52:25.934) Right, right. Like bro culture, drinking, going on elaborate vacations, you know, right? Like that's and that's not brain rot. But like, I think that, you know, if you're spending your time scrolling through random mindless shit all the time, and you're just getting addicted, there's a difference. But there's there's also like, are you spending too much time obsessing about this thing that's not really advancing you in life like

Is it even actually a problem because like a lot of the things that we all argue about and we battle about aren't actual like real issues like like there's never like it That's true, right? The majority of the things that they talk about on that whatever podcast are like these hypotheticals So it's like if there's only four women on the planet left It is okay to sexually assault them to repopulate the world. She literally had that conversation, dude I swear to God like it is they are bat shit

It's going back to that stupid Yellowstone conversation we had, which is like, hey, maybe instead of like, you know, scrolling through Instagram, maybe go read a book. Go read a book and don't just read a book that was written in the last 10 years trying to tell you how to live. Go read a book that's old. No. Instagram is the same level of addictive.

Yeah, there's a cut.

Speaker 1 (53:27.079) So you're not on TikTok? No. Okay, so that's the Instagram.

Pretty much the same thing. Now, what I find, though, is I don't find it the same addictive level. I can get into Instagram and I won't get stuck. With TikTok, I get stuck. Like, I still think that their algorithm is more drug-inducing, dopamine-releasing than what Instagram. Not as a good or a bad statement. It is just what it is. There's one, though, with TikTok, too, is another one that we're starting to see a lot of is you get a ton of Elon, like, posts that mark as Elon, right? They're like Elon Musk, and then it's all just like, just, you

know, mega conservative nonsense. And then you're like, don't show me any more of this, don't show me any more of this. They are pushing it hard. I mean, they're pushing it hard. So like, every time I see one, I say, don't show me this, don't show me similar content or from a creator, the same creator, and it keeps doing it, it keeps doing it, it keeps doing it. That's wild. Like, there feels like there's a very specific thing that they're trying to do to bring more conservativeness to it.

And maybe that's just the way that it works, is every time a new conservative gets in as president, then all of a sudden Facebook's all about free speech and open source, or free speech bullshit, which it's not. And then TikTok just said, fuck it, we're just gonna go lean into it. And so I see way more conservative content now on TikTok than I've ever seen before.

No, that could be that could there could be that could be with that price they paid to stay on the apps. Point well, that's been a good one. Vibe Do it like it. Don't like it. I like you got to play with it. You got to play with it. You to play. You can't. You got it. You got to get in there, though. Right. Got to understand. You got to prompt it. You got to see what's coming back. Check your work.

Speaker 1 (54:54.548) That's a very good point.

Speaker 1 (55:03.214) you

Speaker 1 (55:18.638) Go to Lovable, like everybody, if you haven't done any vibe coding, I recommend you go to Lovable and just put in whatever stupid ass idea you've had that you've been sitting on. And just do it, just play around with it. Because it does excite you, like it gets you going, it gets you thinking about things, and whatever can do that, I think is a positive. Now, you can't necessarily rely on it to deliver you a production platform, but that's okay. So I've got a call next week with somebody who went through this whole process where they built out this thing in Lovable, they hit the 80%, they don't know what to do,

restart and we build it from scratch and you know what I'm gonna tell them? You need to talk to Crafted. I mean honestly.

Well, go download, go download lovable.com. Go to lovable.com. That's just how much I really like your product. Lovable with no, no E.

Hold on, Dev.

Speaker 3 (56:10.806) Well enjoy the call with you.

Is does does the proper spelling of lovable L O V E A B L E I wonder because like I always How would you spell lovable

Leave a comment below. Yeah. How would you spell loveable? L-U-V. No, great product. Go Vibe Code with Lovable today. This has been the Big Cheese AI Podcast. Starring a bunch of crazy lunatics. Brandon, Jacob and Shaw. See you next time.

There it is.