Kevin Williams: There we go. Riverside changed something on us again, Matt. You know, the world changes so fast. And you know what I've discovered in my podcast adventures is that, you know, I may be AI guy, but I am not an A V guy. Like the fact that you and I were fussing around with like, you know, microphones and lights and everything beforehand. I'm like, just just give me my bots, man. Give me my bots. Matt: I have to set my alarm up like twenty minutes before we start this podcast every time because my setup is always like so confusing to figure out how to get the camera on. Like I gotta rearrange the cords. And yesterday we gave this awesome AI presentation at a vistage group of CEOs. And I I couldn't for the li I wanted to record it for the life of me, could not figure out how to get the camera, it was overheating. It was like a complete failure. The the workshop was great. We had great content. It was awesome, but like the camera A V, I completely failed. Kevin Williams: Think we have a ways to go. Yeah. The number of conference rooms I've been in Matt: Yeah. Kevin Williams: and like not been able to connect. And you're always sitting there with really senior people and like fussing around with like whether it's you know, an Apple you know, airdrop or whether it's the right HTMI cable or whatever it is. These are things that AI has not solved for us. Matt: But it also gives us hope because whenever I talk to A V people, they're always like, this is like the easiest stuff in the world. So my point is, is like no matter how easy things become in the world, you know, whether it's technology, AI is gonna make things easier, there is always gonna be somebody that you need to hire to help you out because it's just not within your domain of expertise, even if it is pretty easy. Kevin Williams: So I actually think that's perfect what you just said, because you know, for other people, the types of things that we do and the types of things that a lot of people listening to this show do with AI, I'm not gonna say they feel easy, but they're starting to feel pretty natural as far as new work patterns and primitives and things like that. And then You know, sitting down with with people who haven't been exports exposed to it on a granular basis for the last, you know, four years now, it's still really hard. Like sort of backing up a second. And I know we have a lot of like hobbyist hands-on keyboard type listeners out there, but if you were to put yourself back in like November of last year, it's just That sort of status of complexity still exists. It's just relatively more easy now than it was then. And you know, the industry has a long way to go, and people really want to start solving these problems on their own, but they're just it feels like there are so many impediments in place. And we saw a bunch of sort of movements in the last last couple of weeks as far as making. some of these technologies a a little bit more assailable. And you know, the one that I I really wanted to launch with was you had pinged me last week about lovable in particular and the fact that lovable just raised four hundred million dollars. Yeah. Matt: thirteen billion dollar valuation. Kevin Williams: For a bunch of, you know, twenty something Swedes, I believe they are. Matt: Two years the business is like two years old, you're right. Anton is the CEO from Sweden. He's, I don't know, probably like twenty-two, twenty-three or something. Very talented person, but yeah, incredible amount of money that they've raised in a short period of time. And an incredible valuation. Kevin Williams: So what is lovable, Matt? Like I I I think that that you know, I've hated Unlovable on this show in the past. So there's we're I'm sort of setting this up for a little bit of a debate here, I think. Matt: Yeah, I'm I'm excited because I'm a I'm a big fan of Lovable. Like the main thing is if you go to a barbecue, most people are gonna know what open AI is, right? Like it just we we all live in a bubble, even our audience listening to this, we're all in a bubble, but you have to remember the vast majority of people, like 99% of people on the planet still don't know even a lot of the basics. Like most of our audience has probably heard of Lovable. But a lot of people don't know what lovable is. Like lovable is for those of you who aren't quite sure, it's a it's a platform sort of like codex or cloud code where you can build, but the interface is is just all prompt based. And it's not it it can show you lines of code, but it's not really geared towards showing you lines of code. It's more about just showing you it actually appear while you're building. So its entire niche is more around usability, ease of use, ease of adoption. and so it's it's I think it doesn't get as much attention as Cloud Code and Codecs, especially from folks like you and me, Kevin, because we're a little more technical. But the usability is just is just so much more usable than most of the tools that are out there. Kevin Williams: yeah, so you know Lovable is a platform that an ad an organization can put into place that essentially can control the administrative aspects of apps and also control a lot of the security and trust functionality around apps. Now, that wasn't always the case. you know, I spent a lot of time with Lovable last year. And at the time it was just a a a layer that was sitting on top of the LLMs. And I have a bunch of reasons that we can go into that I developed a pretty deep seated loathing for it at the time. In fact, it cost me a whole weekend that at some point I talked about on the show. But how has Lovable evolved? You you use lovable across your team. Part of this conversation started a couple of weeks ago when you sent me something and I realized, wow, that's a lovable landing page. And we started chatting about it a little bit there. But from an organizational perspective, what does lovable look like in your team? As far as let's just say you have 10 people using it. do you have enough nitty-gritty knowledge of how that of how that works as far as collaboration and how a pretty big, you know, globally diversified team uses the platform? Matt: Yeah, so there's a couple of different value props that it offers, but the one that you're pointing out is the I think its greatest value put prop is that your most of your company, depending on the business you're in, is probably not that technical. And something like Codex Claud Code, Cursor, Vercel, these are tools are too intimidating for most people in the workplace. Lovable is a much more approachable tool. And it can bring if you think about your organization's readiness, AI readiness, lovable is a tool that everyone can adopt with relative ease without being, you know, typically tech savvy. You know, if you're the type of person have people at your company that kind of break computers or have some electromagnetic field that everything always goes wrong when they touch the computer. Lovable is that tool that they can figure out and is not gonna break. So that's like the main thing. It's it's approachability, usability, it's speed to launch. Like you just type something and bam. On first attempt, it gets something out the door really well. It started as just what you had described before, just you know, make some landing pages, make some simple apps, connect the database to a third-party platform, but it didn't have a lot of the control features that an enterprise would use. But in the last year, they've really pushed into the enterprise space. Obviously, there's been a big race between Cloud Code. and open AI and everyone's trying to get enterprise customers. Lovable did the same thing, and they built out a suite of enterprise features to help your organization control the data, the user permissions, what people have access to, helps you control usage, helps you monitor who's who's doing what with the data and what what what tools they're building. So it gives like your your IT department or your engineering department more control over all the non-technical departments that. the shadow IT that exists that, you know, could cause problems if you're not able to control. Now with that, you're gonna have the vendor lock-in, right? You're gonna have, you know, you're playing paying lovable, but you're getting that more granular control. You're also getting the other thing I really love is the ability to kind of route to the model that's most appropriate for the job. On this show, we always talk about burning through our tokens accidentally or not us, but our organization burning through them. Lovable is always doing that back end routing or math to figure out which model is the most appropriate for the for the prompt and what you're trying to build. and and it can Kevin Williams: See the Matt: switch, like easily. Kevin Williams: So that's the part that would just enrage me because I have a pretty good feel for what model that I want to use for one task. And I I have less of a conversion commercial motivation than say lovable does. And while all of the the major platforms are starting to lean into sort of auto modes, a lot of it's out of self-preservation, but the fact is having consistent outputs from a model that you're used to can often give you better products. The other things that drive me absolutely bonkers about lovable, one is the lock-in. Then unless you're very, very careful when you enter a lovable environment. And by the way, this isn't just lovable. So there are a bunch of others in this care in this category. So Replit, Bolt, base 44, the last one, base 44 is the one that really has deep lock-in that drives me nuts. But Let's just say you're you want to create an app or an app set on Lovable. You go in, Lovable has a superbase connection, so to a database. I believe they're also developing their go ahead. So Matt: It's actu it's actually just their own internal database that you can connect to now. Kevin Williams: yeah, okay. Well that is a problem because once your data is established there. Matt: You could do other databases. Kevin Williams: You can do other databases, but most people don't. So you come Matt: Mm-hmm. Kevin Williams: in, you establish yourself, you put all of your data onto their database. And while that database might look like Superbase, getting it off of that database is going to be really, really a problem. And that's totally fine if you're doing like pretty superficial things, like, I'm gonna spin up an event landing page or I'm gonna do a little app that's like a time tracking thing that's, you know, twenty, thirty thousand lines of code or whatever it is. it gets to be a real problem when you get a some sort of an app that's a few million lines of code because you're essentially paying token arbitrage to them. And again, you know, and I'm not gonna like cast shade on an entrepreneurial organization like that that that's why they raised $400 million, right? But as the code base gets bigger, you find that your your your your bills like go through the roof without really having the ability to manage an app. on like a per function basis is what I found. You you you it's super inefficient to to manage a big app. So it's not it's not like a development tool. It is a like a fun tools and toys tool that can do really amazing things. But I caution organizations that that it can be it can be pretty fierce once that gets big as far as a lot of things going on. Matt: Okay, I challenge you, maybe this weekend, check it out again. Because I I have the opposite take. Like, yes, those things that is the brand's perception and that's the perception of the tool. But they're they're very aware of that and they're they're they don't wanna be just an MVP tool or a tool that's a toy for you know, non-technical people in the organization. They want to be the last piece of software that you use. Okay. Which means they want you to be replacing all your SaaS with build builds that you built on Lovable. And not just, again, it's not just like, you know, a cer a small 10% of your organization that's able to use these tools. It's designed to be a tool that can be spread across the entire organization. with a high adoption rate. And I think that's I hear you, Kevin. I'm with you. But I think a lot of a lot of businesses do not have any tech in their DNA. And but they need to become AI ready. And they don't need to do it right this second, but they need to start preparing preparing and they need to imagine they're a business in three to five years where most of their organization is leveraging AI. And part of the way you leverage AI is not just using it as a you know a co-pilot for prompting and helping you brainstorm and you know looking through your inbox and doing some simple things, a lot of departments should be building tools that make them more efficient. Lovable is that platform that makes it easier. Like the reality is just when I when we do training for an organization, we show them a suite of different tools, lovable is always the tool that they grab. I mean there's a few. But lovable is one of the ones they gravitate towards as being an easier, you know, the person who's a little bit older, maybe a secretary or admin or an HR where they're just not that familiar with t tech, they light up and they're like, wow, I can do this. And that aha moment happens. Kevin Williams: So I I mean I I I I I am sympathetic to the argument. I really am. Like it makes sense, particularly since we do granular clo like vibe coding training now. And when you walk into an organization that's really enthusiastic, semi technical, let's just call them like pegged on the curiosity scale and sort of midway on the on like the tech scale, it's pretty horrifying what you have to do. Like again, I Hard fought like lessons that I have personally and my team has from you know a long time of working with this stuff that feels sort of like second nature. But you sit down with somebody who's never interacted with the terminal on their Mac and you tell them to open up the terminal, and then you say, Hey, by the way, you can completely nuke your you know, your file base or all of your emails by going to the token the the the the terminal. Have fun with that. Like hair falls out of like, you know, a CISO or you know, a a technical person. And it's really, really complicated. Like they the the nuts and bolts that are that you have to put into place in order to get Supa base established, which again is is the back end database that we tend to refer, Vercel, which is the front end that we we we we tend to use, using the terminal using local file storage versus cloud file storage, still in most cases needing to keep if you're on a Mac, like keeping it awake while it's running. These are all enormous friction points. And you know the trade-off is what I get for that is I get much more granular control. I really truly own all of the pieces of my stack. And there's nobody else who has their their their like commercial tendrils in it in a way that can really impede the way that that works. And I can I can really build real products. That doesn't mean that I can't have both of these in place, but you there are trade-offs there that that organizations need to keep in mind. But I'll I'll concede that in a lot of cases, today's lovable, which I do owe it to myself to spend a decent amount of time like actually getting in there and playing with it again. But I felt so burned by it last year that it really left a pretty negative taste in my mouth that I'm I'm gonna I'm gonna have to slowly acclimatize myself to. Matt: I think of it this way, is like what what's the gap between the frontier models and lovable? Because lovable is you're like you said, they're arbitraging the models, right? They're essentially plugging into all the main frontier models and they're gonna they have their own algorithm for routing which one model they think is best for the particular task or prompt that you've entered. and then they're charging you little bit on top. But like the mo the frontier models have gone so good since probably the last time you used it, like just inherently level will be that much better. And Like I just, it can't be that much of a difference if they're all using the frontier models underneath, there can't be that much of a discrepancy in quality between using lovable or using the frontier model directly. Kevin Williams: So I'm not after quality. Quality is not the issue. It is actually its cost in particular. It the lock-in does indeed bug me that okay, now, okay, my organization has 48 different apps that we've built internally. That's just running like our own things. If all of those were baked into Lovable and Lovable decided to change their business model, I mean, this is a deeply cynical view, but like it's not like Google that's out there monetizing you in other ways and generally giving you their their tools for free. Like there is a direct commercial motivation here. And once they have you locked in, even if you have done the right things as far as your data structures, getting your information out of there, it it's really truly not going to be portable. So again, good for them and their $400 million. And as a retail investor, like if they IPO'd, wow, okay, this is probably a pretty good bet. But if you want to be an agile organization and Not everybody does. Like like I don't actually I don't think it's a good idea for a lot of people to be like building like completely random things. So, you know, there's there's there's a there's a giant swath of the market where this makes a ton of sense. Matt: Okay, what do you think about this? I have two more points. One one point is about agenda capabilities of lovable, which you're probably also once you dive in again you'll see. but the first point is like the portability, like you can export the code and if you if your leadership at your company mandates that you don't use lovables, in in in built in database, you use something else which you can hook up. yes, that decreases the usability, but it alleviates some of the concerns you have about portability. So is it there feels like there's a little bit of a middle ground? Like lovable has the data controls for your organization, it has the permissioning, it has the way for you to prevent a lot of shadow IT from happening across the org. It has a higher adoption and usability rate. And if you are worried about portability and lock-in, there are ways to kind of like instruct your team to use it that allow you to export the code or keep your data in a different different database so migration can be easier in the future. Kevin Williams: Okay, so there's some building blocks there that people need to keep in mind. one is you can just like pick up your suitcase and go over to Lovable Land and say, okay, we're gonna do this and we're gonna commit to this, and the cost looks manageable, and we're just not gonna think about it anymore because that's that's easy. if you want to preserve some optionality, then you need to have a couple of different pieces in place. One is that data piece and understanding how that works because Just having access to the data doesn't mean that it's going to be easy to transfer. And that's the nightmare that I went through as it was it was incredibly difficult to reconstruct the way that Lovable was looking at the databases, even though I had access to the code for a lot of technical reasons. So okay, yes, having either backup access or direct access to your own data, that's smart. That's like, you know, your your your outputs, your back end, like you want portability there. They don't make it super easy from poking around with it. Like, like you can do it, but it's not it's obvious that they don't really want you to for again commercial reasons. The other piece here is GitHub and like mandating that you have a a proper GitHub practice in your organization such that your code bases are all being backed up to GitHub. Again, they don't make that like completely obvious. But that's a pretty click click easy thing to do. So in that case, your code is backed up to GitHub and your data is backed up to an accessible database in some way. So that I mean, what if they were, you know, e Lovell's doing fine. I don't think they're gonna go out business anytime soon, but like it would be a complete nightmare for them to go poof. Like they have some sort of, you know, FTX type like crash and like all of a sudden they go under, like, my gosh. Like Matt: I know, but that argument could Kevin Williams: Yeah. Matt: be applied to Salesforce or any other, you know, your ERP. Kevin Williams: No, we can no, come on. This is a two year old company that's run by 25 year olds. So it's this is not Mark Pettyov. Matt: I mean man, they almost raised a billion dollars in total over the last two years. I think there's something there. Kevin Williams: So I I you know, I we have clients that are that are enterprise level clients who are like, absolutely not. I am not using whatever little startup that you just put in front of me. I've been burned before. I will sort of like gut and you know, squint and use Salesforce or HubSpot or whatever their current you know, kind of the craptastic tools are out there. Th this is not a non zero risk. So like so Matt: I agree, not definitely not not in zero. Kevin Williams: so again, it's It's neat, and if you're trying to get going, you you listen to us and you're like, man, that sounds really complicated. Well, you're not wrong. Okay. It is not not complicated. So I do think that there are a lot of other tools that are coming on the horizon, too, that are showing showing us the way to the future here. And I think that us in a year, we're gonna look back at the sort of crazy town stuff that you know. I was last night. It was like I Last night I was building something and Matt: Keep going, Kevin. I'm here. my camera died. Kevin Williams: Last night yeah, Matt Matt just had another A V issue. So he he he up and vanished on me. so you know, I'm I'm building something and it it was it was Claud code was being a little bit persnickety and then Claude dropped and then I had to reconnect this and this side and it took a long time and it's so kludgy right now. And we're entering a phase of consumer accessibility where these big brands are starting to figure it out more or less. And lovable is a really good example of this. I've been very fond of Replit as far as usability, but again, I haven't used it very much in in in a while because we went full in on cloud code. But I think that that that that you have it right and they have it right. That making this just so much more accessible to sort of normy people, such that Sally in accounting can have access to data sets that her organization feels that she should have access to and be able to build what she wants to build in there and not be like bound up over whether or not she's going to accidentally overwrite all of her local data on her Mac. Like that's just a better experience. And Matt: Yeah, totally. I like one one more thing I'll add here, Kevin. I'd like to get your take on, is we all know that data centralization is real or data management is very important for like increasing the agentic capabilities of your organization. So you one thing that I think is interesting, and this is more gonna be we're leaving like the debate of, you know, is lovable, you know, the the pros and cons of lovable and kind of projecting out into the future. I think there is an advantage of having all your data in kind of one pl ecosystem or platform when it comes to making agents more capable within your organization. Now, I again we're talking about a certain type of organization. We're let's narrow our conversation to that org that's less tech savvy, has a lot a lot of non-technical personnel. building in some of the other tools is just too like too too hard, too intimidating. But so for those organizations that want to pursue eight building agents, not just little mini apps, you know, lovable is going to be releasing a bunch of features that make building agents, if you have all your data within lovable or like a lovable or accessible environment, building agents becomes easier. and I think that's Kevin Williams: Yeah. Matt: like a challenge. Like, what do you think? Kevin Williams: I think that that's not necessary. that that that people get really bound up over this. I have CIO type conversations all the time. And they they they like get all warm and fuzzy about the data lake that they can build, and they're gonna pull Matt: Yeah. Kevin Williams: all of these different things together into their data lake. And you know, they've always wanted to do this. And you know what? Connecting these various APIs, particularly if they are controllable on the admin level, is Pretty much trivial at this point. And you it's not a big deal to bounce off of your CRM and the data store that's there, and then grab information from your ERP and use the information that's there. And put and of particular interest to me, and this is a little techno-dork, is the fact that the Data being stored, not in a data lake that is like heavily managed. And don't get me wrong, they're like big organizations do need data lakes, and it's like this big deal and it's millions of dollars, and yada yada yada. But more like mere mortal organizations. What you miss in that conversation, or what lovable is this isn't their core expertise, is the schema that's associated with that data. Like CRM data looks a very specific way. Project management data looks a very specific way. And if you just try and dump all of that, and dump is not a fair term, but centralize all of that into like some data store that lovable has access to, that doesn't solve the schema problem at all. And lovable is an is a vibe coding app that is democratizing access to building this stuff. It's not a data lake builder. And maybe it will be someday. But I don't necessarily see the advantage and you're actually ending up you could be costing yourself some functionality because the data ends up looking different than the LLM is expecting, basically. Matt: Okay. I think that's fair. I think I think the maybe like shifting the conversation a little bit is like agentic building agents is like an area that a lot of organizations want are are interested in. And like we've talked a lot about vibe coding, and vibe coding is mostly around building interfaces for our audience. Like this is about building, you know, maybe a little tool for your team internally, maybe a little tool for your vendors or your suppliers, tools within different departments is not maybe something that touches the entire organization. But I think a lot of people are still not totally sure how to be building agents for themselves personally, or how to like educate their entire company on what tools are out there to build a. Agents and grokbot just came out, which is like kind of interesting how XAI just started to catch up finally on this agentic AI front. so yeah, maybe maybe lovable won't won't be that tool that helps, you know, maybe that data play isn't as is important for helping organizations build agents. but yeah, what what did you did you play with grokbot at all? Like, what's your view on this agentic AI with grokbot and like how organizations are trying to figure out how? They can be building agents not just within their tech team, but also within these non-technical departments, like you know, the citizen developers or vibe coding apps. Kevin Williams: So, okay, you just sort of dropped that on us. So XAI is owned by SpaceX, which is owned by, well, you know, founded by, well, not founded, but you know, controlled by Elon Musk. And they bought Cursor a few months ago. And Cursor is probably the dominant IDE, which is a coding environment that essentially facilitates using AI tools within more. I don't want to say traditional, but within development environments. And that was a pretty decent coup because Cursor was way, way ahead as far as understanding how LLMs and coding work. And you know, Elon's not afraid of spending money. So they leaned really, really heavily into XAI. And and I am not a fan of Grokbot, of Grok in general. I have not been. I don't, I don't, I'm not really an X user. and I never really found a ton of utility in it, and it never really scored particularly well. And then all of a sudden they had new models drop that were absolutely exceptional, and like overindexing on the charts. So basically in six months, it looks like we went from okay, in twelve twelve months ago, you had three major frontier models. You had open AI. You had anthropic and you had Gemini. And then there's another topic here. Gemini has really floundered. And although I'm a big believer in Google and they have this is kind of their game to lose, like their models have just not kept up and their features have not really kept up. So even in our own documentation, we've only been talking about two frontier models, Anthropic and Open AI. And now all of a sudden XAI has popped on the scene and it's like a real player. And the first manifestation of this that dropped in the last week were Grokbots. And Grokbots do require a a pro cursor subscription at the moment, which is $200 a month. So my gosh, $200 a month. but it does add up. I mean, you spend thousands of dollars a month on on AI subscriptions. And the deal here is they're essentially an evolution of OpenClaw. So this show has talked a lot about open claw when it dropped at the beginning part of 2026. And this idea of having sort of a self-morphing, self-guiding set of agents that can do things and occasionally like nuke all your email. open claw was deeply flawed, deeply expensive, but also deeply, deeply cool as far as showing us the way that this stuff was going to work. Then it evolved into Hermes. Which was a more refined version of OpenClaw and the ability to connect to a whole bunch of different data sources and do neat things. What Crockbot has done is it's the same conversation with Lovable, is they've made it really, really easy to use these bots. And their version of it is they're basically, we don't need to get overly techie, but they're basically sir spinning up. almost an individual computer entity for your organization and then compartmentalizing that such that it can operate in the cloud and go off and do things on your behalf. log into apps, grab files, it can look at code, it can do all of these things. And you know, there's a lot of hate going around on you know the AI, TikTok and you know the influencer scene about it because You know, fact is you can do almost everything that it does using existing tools that are out there, but it makes it easy. It makes it click, click, easy to set up a little bot that's gonna go out and do. you know, a friend of mine runs a t-shirt company and he's having it do social scanning for concepts. it has a very strong image capability. So it's immediately mocking up whatever the latest trend on t-shirts are. And then it's circling around and dropping that in his Slack channel. And then from there his designer approves it. And then he uses another bot that moves it over to production. Like it's pretty phenomenal. And this is a pretty techie person, but like it is pretty phenomenal to see what people are starting to build with these things. I don't have any hands on keyboard time. So I can't, I can't like I don't have a great grasp on how I would use it in my organization. And I'm pretty slammed right now. So I'm I'm sort of watching this one develop. But what have what have you heard or what have you seen, if anything? Matt: Same same thing. Lots of people are excited. generally I I've been through so many hype cycles that I'm trying not to get overly excited. I'm sort of taking the approach of like, let's give it some time, let's see what happens. Like Open Claw kinda happened this way. You know, we we ran this giant conference in Boston. We had like 300 people show up to an open claw event, but a lot of the enthusiasm has faded around Open Claw. And I it just a lot of these things. And then, you know, the next thing comes out and it grabs everyone's attention. And and Open Claw had some serious, like definitely very difficult for for people to use you know, I mean, y you didn't have to be overly technical to get it to work. But I think for the vast majority of people, it was again too intimidating. and so anything that's more usable by the masses, I think just gets more attention, has more lakes to stand on, has more potential. But I sort of want to sit back and see. but I like these tools. I like these tools that are like focus on the usability versus the cost or the the capabilities or the quality. Cause I just think getting it in the hands of, you know, like 10% of your org versus 90% of your. org is so much more exponentially impactful than you know maybe you pay like 25% more in cost or maybe you have you know 20% less capabilities or fidelity with certain types of tasks. I think those are marginal things to worry about versus like the the bigger exponential impact is how many people are using and leveraging these tools to make your overall company more productive. So I'm excited for it but I myself as well haven't delve in kind of waiting to see What the internet chatter is all about, and if people start pinging me in my network to say, Matt, you really got to take a look at this. So I like to give it a couple weeks. Kevin Williams: So, I mean, this these still require a learning curve, right? and just sprinkling lovable licenses or a grockbot or I guess cursor licenses across your organization is not going to result in much unless there's a common understanding of how you're supposed to use these tools. Now, I think that there's a bet that the tools themselves will help you figure that out. Like, hey, this is my role. Like have a conversation with it and you know I imagine that GrockBot's gonna come back and be like, I could totally help you do this. I can help you do that. And now you're kind of off to the races. But I would be a little afraid. Okay, so this is on the on the continuum of most regimented is like IDE-based development structures using things like cursor, traditional dev approaches to frame things up and then very gradually expand from there. That is the right answer for like product integration for bigger companies who are trying to figure out how to use AI AI. The next step on the continuum is you know codec or cloud code where you're doing similar work but it's a little bit faster and loose and it's a little bit more democratized such that your citizen coders can get in there and start building. But if it's well constructed again it has pretty decent like guardrails. Then you have the lovables of the world, which are a little nerfed, as we talked about earlier. And I know lovable lover lovers are going to push back on that, but I I'm going to maintain that that you you just don't have quite as much flexibility, but you have this nice like garden that you can use. And now you're actually bridging away from that a little bit into terms of like the Grockbots and or Hermes or whatever, where having centralized control over those at this moment in time, I don't think it exists at all. So now you're gonna have this like little army of bots that are doing things in your organization. And you know, if you're like super entrepreneurial and you're signing up for that, I mean, great. But again, like CIOs listening are like twitching. Like you y this little thing is gonna spin up and it's gonna have access to things and it's gonna have credentials and it's gonna log in. my gosh. So It's a lot of the Wild West that's that that's happening with these and I don't think it's gonna happen like by accident quite yet. Soon, but not quite yet. Matt: What I was saying is like a lot these arguments are the same sort of arguments you have with vibe coding, like you know, all these people are gonna spin up apps and they're all gonna have different the wrong permissioning and data usage and everything. Like what what's more scary to you? Like a bunch of vibe coders coding up internal tools or tools that have access to different databases, or a bunch of agents that they code that can go wreak havoc in other ways. Personally, I think the agents can be more destructive in the shadow IT space. And this is like again goes back to Yes, vendor lock-in is a problem, but if you get a tool that can help you manage the the risks around shadow IT, I think that's the the stakes are higher with the Gentec AI versus just vibe coding the. Kevin Williams: So these are things that people didn't talk about. Like outside of technical like corridors, our audience are like, you know, operations executives and marketing and CEOs and whatever. And, you know, basically other people would deal with security and things like that. You buy an app, you sign up, it's got terms of service, you do your thing, you feel good. And you're torn between these new toys and tools. Being an entrepreneurial amplifier, like, okay, I'm a I'm a high trust driven organization. I'll take some risk on the edges. But then you walk into an organization that's like a normie organization that has 200 people in it, and it's like the the people don't understand this stuff either. So you can be exposing yourself to a lot of risk. And you know, I I I'm a little sympathetic for the general counsels and the CISOs of the world who are like, guys, we we just can't be doing this. Like you're creating vulnerabilities left, right, and center. So just I lovable probably rounds off the edges of that. And there are other enterprise tools that you can put into place that that do a lot of the governance aspects and make sure that your whatever you're putting out there, particularly if it's client focused, have you know, some guardrails in place. But the agents, okay, the agents are different. That we talk about probabilistic versus deterministic on the show all the time. You're gonna build an app and it's gonna be a time tracker app. Yeah, okay, it might have some probabilistic elements in there, in that it looks at your calendar and it like makes sense of it or whatever. Agents make decisions that is different than building an app that does like a thing using AI. So when you launch an agent. Matt: Yeah, like the placement of the bubble of the button. Like that is probabilistic. Like, okay, the buttons generally the hamburger's best on the right, but some people put a side panel on the left, you know, like there's different it's not it doesn't break anything, you know, like I mean it could break something. Okay, let me rephrase. But like every step of an agent is making a decision. Is that your point, Kevin? Kevin Williams: Well, it can if they're designed that way or if they're casually designed that way. That that's the yeah, we talked about like the incursions or the sandbox breakouts. And it's because the agents were given the latitude to basically make decisions. And, you know, as much as I like Elon's products, I don't necessarily trust that there's a lot of like governance or guardrails in place. and it Sounds to me like it would be pretty easy to accidentally the same guy was like, Yeah, I just had it like log into my bill dot com account and it reconciled X, Y, and Z. And I'm like, my gosh, you you just let an agent just like log into your receivables account and do stuff. I mean that's that's that's pretty aggressive in my opinion. Matt: Yeah, I mean even like giving it access to email, I think it's still like I I'm not sure I like condone even giving some of these agents like right cap right or s well drafting capabilities, but sending capabilities and your email permission. That's a bit scary. so and Yeah. Kevin Williams: Yeah. Yeah. So it's early days. It's early days. Like it it really is. As much as man, I've you I've I've gained more gray hair in the last four years, like just like leaning into this and understanding it and bouncing around, but we're still in the early innings as far as the way that this is going to play out. And it's entirely new structures that you know, organizations aren't ready for. So okay. Take one of our, okay. I mean, we're exceptional organizations because obviously we live and breathe this stuff. So we're making AI first conversations. But take an organization that's like, you know, I don't know, just sort of a normal industrial company or whatever that has a hundred front office people and they don't have tech infrastructure. They have some MSP that's managing their networks and their computers, and maybe they have like Joe, the web developer. And otherwise, Like there's nobody in that organization. Like I always say, there's somebody in there. And yeah, okay, there's somebody in there that's building like some cool golf app at night or whatever the heck it is, maybe. But I'm finding that sometimes there just isn't. And if you don't have any sort of technical foundation and you don't have any sort of controls, and your investors, your stakeholders are telling you, well, you need to lean into this, you need to lean into that. If you don't understand it, at least to some moderate degree, you're gonna get in trouble. Matt: Definitely. And I think agentic AI will get you into more trouble than the vibe coding phase that we went through. One thing I'll say though is I am glad that like XAI is getting back in the game because, like you said before, it feels like Gemini is kind of leaving the game. Right. And I don't think a dualopoly between just open AI and anthropic would be good just for this industry or our economy, especially if we like, you know, the government steps in and says nobody's allowed to use the Chinese models for, you know, security reasons, that makes sense. But it for competition within a capitalistic society. society, I don't think it's a good idea just to have two models competing. And it kind of feels like Gemini is taking a step back. I don't know for you if you agree, but y like you said before, they just haven't kept up the way we would have expected them to keep up. They've just lost a bunch of staff for for our audience. This was big news in our world. Three top people within their AI lab decided to leave Google. and I you know, some people are interpreting this as a sign that Google is gonna focus on infrastructure and not focus on trying to be open AI and anthropic. which I think would be bad. Like I think competition is good and we want these all these these companies competing, especially Google, because Google Has all the data. Like, if anybody should be really like positioned to win this, I would have thought Google had like such an upper hand. Kevin Williams: So not an investment show, but I have been bullish on Google for a long time because they again, as I said earlier in this show, it's this is sort of their game to lose. They have the data, they have the money, they have the infrastructure, and they it does feel like they have pivoted some of their efforts towards just being a provider of that infrastructure. we we promised in this show that we don't go into like chipsets, but they have a they have a differentiated chipset. They offer this more cost effective and they're renting out a lot of that capability. I'm not counting them out. I mean, you've got Sergey Brin has has basically taken the helm again. He basically said, you know, this is the most exciting time in development ever. This is way too much fun. The yacht's too boring. So I'm back in and you had Demis Asabas left and they immediately moved the research group away from London. where Demisabis would resided. So that's where those research research researchers were. And they immediately moved it back to the, you know, Google Mothership. And Bryn is like personally focused on it. I just, I just wouldn't count them out. and I I I there's a lot of potential there. And then then, you know, we we talk in this show a lot about large amount models, but something that's being a little bit lost in the weeds, I think, is the dominance of Android like phones that are out there globally and the capabilities around small model development as as far as like small models that can sit on a device and process Matt: Yeah. Kevin Williams: on a device. maybe that's the direction they'll go. And I truly don't know. Like small models, I've own almost no expertise in small models and how that that actually would work. but I'm just not counting them out yet. It it it and I I'm also I feel like the older I get, the c sort of the leck the less deference I have to like Wall Street and what Wall Street does. And it's like I I it's not they they're very, very smart, but they're sort of chasing leading indicators and yeah, quarter on quarter indicators. And I I do believe Google's the sort of organization that plans for the long term. So yeah, not counting them out, but wholeheartedly agree with you. that and I I am a little I'm a little pleased about this, but everybody was rattling cages just like three, four months ago about, the AI subsidy era and everybody's gonna have to jack up prices. And yeah, prices have definitely gone up. It is I spent a lot of money on tokens in the last month. But there's so much competition out there. And it's not, you know, every time Anthropic tries to raise its prices, like open AI is like, but, but, but Codex is great and it's cheaper. So come over here. And then they lose users and like they haven't been able to switch into a mode of of like profit harvesting that I think that they were hoping for. And with Google being out there as a potential spoiler, Google, you know, again, it doesn't have to be a frontier model, it just has to be a very good model that's cost effective. And then you have the Chinese models, and now you have XAI. I mean, this is nothing but good. For us as operators and consumers of tokens. It may not be good for economics in general and how these business models play out, but for folks like us who are playing in the application layer, this is great. Matt: Yeah, I think I agree. I think this is sort of a Wall Street narrative that's gonna, you know, just makes like for a good story and a lot of eyeballs and intention. But I think just using my my gut instincts, this is too big of an opportunity and Google has too many stakes in the iron and too fire and too many too many advantages over the other players to just give up this easily. So I I I highly doubt it. I think this is just a fun headline for the week for people to pontificate about, but I I highly doubt they're actually out of the game. those switch it like all the things you said, maybe they won't compete like you know on every single metric. Maybe it'll take a slightly different value prop or or a different direction, but they're still gonna be in the game. Kevin Williams: Yeah. And the yeah, they have pole position, they're there. yeah. We're it's still going to be a thing. I it's funny, I I feel like I I cross between platforms all the time. I have clients that that have me working in w in Microsoft, I have clients that I'm in I'm in Google, so I I have Copilot and I have Gemini and they are all all starting to blur together as you know, they're not dumb. They're paying attention to what other organizations are doing. They're paying attention to what like Anthropic is doing on top of their models. And it's not rocket science necessarily to replicate features. So the features are Matt: Mm-hmm. Kevin Williams: easy. The models themselves are really complicated. And I think we're we're we're we are in a like this golden age of like all the features are going to be everywhere and everything is going to start working a lot cleaner in the next six months. Matt: Yeah, and a lot of orgs are like, I'm already on Google, I get Gemini for free, or I'm already on Microsoft and my plan gives me Copilot for free. That's good enough, you know? Exactly like you said. So Kevin Williams: Good enough. It's good enough. All right, sir. I think that's the lot going on. I think we both owe it to ourselves to go take a little gander at Grokbot and see what else we learn. It is, it is truly polarizing in the in like the AI influencer sphere. Like some people just loathe it. And others are like, this is the best thing since sliced bread. And it's like part of some of it also, I look at it, I think, some of these people are getting paid. Like there's there's there's some Teola going on in here Matt: Yeah, of course. It's hard to know what's real. Yeah. Kevin Williams: somewhere for just like the level of enthusiasm. but we'll see. We'll see how it plays out. And I'm not sure I'm gonna be able to do it this week, but I need to look at Lovable too. Matt: Yeah, check check out lovable. I think we'll have continue the Grockbach convo. And then next week also for our fans out there, we'll t dive into this d AI detection. how how Kevin Williams: yes. Matt: all this stuff is being you know, on social media platforms and other places where you work, you're gonna start being able to see like this was, you know, written by AI or this is fifty By AI. So a lot of this transparency stuff is gonna be coming out that'll be exciting to talk about next week. Yeah. Kevin Williams: It's raining slop. Raining slop. All right, Matt. Good show as always. Talk to you next week. Matt: Yeah. Bye bye.