The Best AI Coding Tools to Use in 2025 | Colin Matthews

It's kind of funny like as time has gone on the tools have seemingly converged in their functionality. So you know at one point in time like you know only one of them had a Figma import or sum base in but now they all can really support all the same features. And so this is funny cuz Google has the best coding model right now but I guess their tool is underwhelming. One thing you'll probably notice is that all VZero applications kind of look similar or the same and it's kind of not that hard to tell when an application was built with VZ. This is Firebase Studio. Um, so there's nothing. I can't move. There's a red box over here. This is it. This is the whole thing. Uh, so yeah, there's I can't do anything. Upload has like this kind of intro screen. It has a mute and unmute button. I couldn't figure out if there's actually any audio playing to be honest. I couldn't get it to work, but theoretically I guess it's there. Uh, controls and so on. And then we have like this thing here. So this to me was actually incredibly impressive. as of April 2025, what's the current best tool for text to prototype? Um, okay, welcome everyone. My guest today is Colin Matthews. Colin is a friend of mine who has taught over a thousand students on how to prototype apps with AI. And in this interview, we're going to discuss the best AI coding tools that you should use in 2025 by use case from simple front-end prototypes to full production apps. So, welcome Colin. Yeah, I would be here. It's my second time and I'm sure it'll be a good one. Okay, so let's make this like super practical, man. Like um so I thought about like just trying to get us to pick the best tool overall, but it's kind of hard. It really depends on the use case, right? Like we talked about. So I came up with like five different use cases. So, so it's like uh simple text to like front end prototype using a Figma design convert to prototype building a JavaScript game uh building a full stack so like front end backend personal app and then uh you know nine of us are professional engineers but like professional engineering um and and uh I think you actually went through the effort to actually like uh pull up demos for each of these use cases. Yeah. So why don't we talk about them one by one? So, I'm going to stop sharing and then maybe you can share to show your demos. Sure, that sounds great. Yeah. So, we're going to start with the text to prototype, right? Um Yeah. Cool. So, for the text to prototype one, you can see that we have a couple different examples. Uh it's all going to be the same prompt. So, you can see at the top here, I have clone LinkedIn. We're going to use the same prompt for all of them. And this is all like the first shot, meaning I didn't change anything. I didn't iterate over time. I just kind of let it the tool do its thing. Usually obviously if you're prototyping something you might put in more than one prompt and then yeah we have bolt vzero lovable and replet for this first set and then we kind of have different tools as we go through the different use cases. So this one is bolt right? Yeah. Yeah. That's right. This is bolt and um yeah we'll just kind of iterate through these real quick. So honestly this one's a little bit hard to judge. I'll just flip through them first and we can talk about which one's best afterwards. So this is Bolt. This is V0. This is lovable. And then this one is replet. And okay, obviously a little bit hard to tell which one is best. They all implemented something that looks like LinkedIn, right? Um there is a one kind of big difference here with Replet being again that it did both the client and the server side and so like it actually produced more code. Um that could be a good thing or a bad thing because if you want like just the UI having more code actually means you have more code to maintain even in the prototype, right? Like you have to make sure that all the changes work. So uh for just visual prototyping like I usually don't use replet just for that purpose but for the rest of them bold v0ero and lovable this is just client side code visual prototypes that's it got it yeah it looks like the design is pretty similar with each other uh yeah it looks like they look pretty similar I mean so replet actually uh has a database or yeah if we take a look at the code here you can see that we have like the client side code so that's what we're seeing in the UI and then we also have some server side code. So, we have API routes. Uh we could theoretically make updates or make changes. And then right now, I didn't actually add a database to this project, but you know, adding a database is as simple as clicking the database button, adding a database, and then you're good to go. So, yeah, it has way more complexity than the other ones here. Okay. And I do notice that for Vzero, they had a slightly different design, I guess. Yeah. Yeah, this is a great call out. So, Vzero, as we kind of mentioned, has a lot of defaults when it comes to styling. So they're going to use Shadan UI, which is a UI library. They're always going to use Nex. js, which is kind of like a development framework. And so one thing you'll probably notice is that all VZero applications kind of look similar or the same. Um, and it's kind of not that hard to tell when an application was built with VZO. I think that's less true for the other ones. Like both lovable uh and Replet, they don't have such like distinct styling all the time. And again, that could be a good thing or a bad thing. like if you like this look then it's great because it's going to be consistent. If you don't like it then obviously it's can be a little bit troublesome to kind of get something that looks more custom. Um yeah, personally like you know after prototyping a couple times I'm kind of tired of the chassis and UI thing. Like I wish there was option to pick different ones. Yeah. Yeah, agreed. Yeah, usually in Bolt. So, I use Bolt relatively frequently and um I'll have it just create components for me from scratch and it'll still implement some like base UI libraries sometimes for like the buttons and stuff like that, but I find it easier to kind of like do custom stuff within Bolt. Uh you know, they actually do support not just Nex. js, like if you want to use different uh libraries or systems, you can do all that within Bolt. So, it's a little bit more customizable. And uh let's take a look at Lovable. Like one thing I learned about Lovable I mean this looks I just think Bolt's UI looks a little bit better than this right for this specific use case. Yeah. But uh another thing about Lovable is like I realize that like you can't actually change the code like directly like you have to hook up to GitHub or something. That's right. Yeah. So they have this kind of code preview here. Um and you can see at the top it says read only if you want to update the code set setup sync with GitHub. I believe they recently shipped the feature for maybe it's in beta or something like that where they are trying to make it so you can edit it. Um yeah, but yeah, I would agree that like there are actually lots of times where like looking at the code this kind of I mean you know Cali this kind of sucks, right? Like if I want to change some string or something like Yeah, exactly. Like if we just go to like I don't know um the index page or let me see if I can find something here. Like if I want to change this text here, you know, grow an exchange with your network. I can't I literally can't change it right now. Uh which is really annoying, right? And you can kind of prompt your way through that. They do have the visual selector, right? So you can click edit and kind of go through that whole process. But yeah, I think you'll find um yeah, go ahead. What is

that? That that is actually pretty unique to Levible, right? That edit button. This episode is brought to you by Vanta. Whether you're a starter founder navigating your first audit or a seasoned security professional scaling your GRC program, proving your commitment to security has never been more critical or more complex. That's where Vanta comes in. Businesses use Vanta to build trust by automating compliance for in demand frameworks like SOCK 2, ISO2701, HIPPA, and more. And with automation AI throughout the platform, you can proactively manage vendor risk and complete security questionnaires up to five times faster, getting valuable time back. Venta not only saves you time, it also saves you money. A new IDC white paper found that Vanta customers achieve $535,000 per year in benefits and the platform pays for itself in just three months. Go to vanta. com/peter to meet with the Vanta expert about your business needs. That's vanta. com/peter. Yeah, it's like a visual edit thing. Yeah. So, exactly. We can kind of click let's say it's like John Doe here. We can just change the text here to myself like so. That's pretty unique um that type of editing. But in terms of like the UI selection, they actually now again I kind of mentioned this, they all converge. They all have this little selector guy now. So if I refresh this one here, we should be able to use the inspector to select UI elements like this. We just can't change the text. We got it's a prompt based on it. Okay. But tell you can probably tell AI to change it for you or something. Yeah, exactly. Like uh like so. Right. So slightly different, but honestly like for most use cases, it's pretty much the same thing. Got it. Okay. So, so what's the verdict, man? What's the current as of April 2025? What's the current best tool for text to prototype? Yeah. Personally, I still think it's Bolt. Um I think Bolt gives you the most flexibility. I think Bolt kind of has some good defaults. And then when you kind of want to go off in your own direction, Bolt doesn't really stop you from doing that. Yeah, I think so. I also think like uh the speed actually really matters for prototyping, right? because you want to get the image fast and like show and make iterations fast. So speed is probably like one of the most important things when it comes to prototyping for me. Yeah. Yeah, for sure. Especially because like you type something in and then it does it not quite how you thought and then you have to go backwards or change it or something, right? And so yeah, uh if you have to wait like 30 seconds between or a minute between every single prompt, it's not a great experience. Got it.

Okay. So now let's talk about Figma to prototype. Sure. I mean, Figma should build this stuff themselves, but like in the meantime, let's talk about which tool is the best. Yeah, for sure. I think there's a Figma uh comp coming up soon, right? In a month or so. Uh so, it'll be interesting to see if they release something for this. But, uh yeah, this is my Figma mock here. This is actually a public template uh that I just grabbed off the Figma marketplace. So, if anyone else is interested in kind of replicating this, you can find the same thing if you'd like. Uh but it's pretty straightforward, right? is literally just um a table, some leftnav, search bar, not nothing too complicated. And so um yeah, up at the top here we have a couple different tools. So I did both Figma imports as well as screenshots just to show the differences. So we have bolt vz and lovable for the Figma imports. And again, most tools at this point support Figma imports. At a certain point in time, it was only, you know, some of them, but now they're pretty consistent that they all support that. Yeah. Cool. Um so we'll start with bolt. Here is our bolt copy. And it's closeish. Uh there's some things that aren't quite right like you know like this little padding or you know shadow effect here is kind of strange. There's like no padding here on employees. Like it's obviously not responsive right because you can see it going off to the far right hand side. So yeah, it's not responsive. We also have like these little Apple indicators inside the mock which is part of the design. So, like I guess that makes sense, but it's also somewhat strange. Uh, and yeah, like as you scroll down, you can see like the left nav doesn't extend to the bottom of the page, and we have all this bottom padding here for some reason. Got it. So, got it. It's okay. Uh, I don't know. What do you think? Uh, I mean, let's look at the other ones. Let's look at V. Sure. Yeah. So, Vzer's uh actually works a little bit differently. And this is one interesting thing about Figma imports is Bolt and Lovable both partnered with third party companies. V 0ero does it themselves, but as far as I can tell, they do it based on a screenshot. And so they all work differently. So this is the VZ1. I would say like this is the best UI, but also no doesn't really look anything like the mock. Uh yeah, it's like chassis. I can't Yeah, exactly. So it's like vzeroified almost, right? Like yeah, it all it's all functional. It's responsive. Um, you know, like we have all the UI elements making sense, but it's really not anywhere close to what we started with. Got it. Yeah. And then last one is lovable. Uh, lovable seem to struggle with this one a little bit. So, you can see like we have some weird leftnav stuff going on here. We only have one record in our table. And then, um, some of the other stuff seemed okay up the top here, but yeah, honestly, not fantastic either. Um, interesting. Yeah. So, so yeah, I mean looking at all three of these things, I probably have to go with bold again. I mean, yeah, the closest to the design. Yeah, I would agree. Yeah. Um, the only thing with the bolt one that I will add is, and again, this is a bit more technical, but the code itself is a little bit strange. So, you can see they actually have like a screen section and then they have sections within the mockup wrap. All of this content is actually mimicking the structure of the um, Figma file. As you can see, the layers. Yeah. Yeah. Exactly. So, it's a little bit strange because usually what you want to have for any type of like front-end code is you want to break it up into reusable components, right? So, like we'd want the left nav to be its own thing, the search bar, the table, all that good stuff. And that doesn't always happen when you have the Figma import process. So, that's just a heads up that like sometimes you can get a easy first step with the import, but then continuing down the path of like extending it can still be tricky. I mean, and also like most Figma files that I've seen like don't have clear labels for the layers. It's just like one, two, three, or like some rand random string. So, so I'm not sure how much they take that to make the actual files. Yeah. Yeah. So, definitely something to pay attention to. Okay. Yeah. So, um let's talk about the screenshots now. Yeah. Exactly. So, this is actually like a somewhat viable approach. Obviously, again, it doesn't really look that much like the mock, but if you think about just took a screenshot of something, threw it in, and you want to test an idea, it's not awful. Uh, so this is the bolt one, and then here's the level one. I actually think the level came out way better from the screenshot than it did from the Figma mock. Yeah, a lot better. Yeah. Which is kind of interesting. Obviously, from this, I would say like lovable has bolt beat on the screenshot here. Um, but uh yeah, kind of it's worth trying both approaches. like just try with a screenshot and see how it goes. Um, and then the last one I have here is this other tool we actually haven't talked about so far with magic patterns. Uh, this is a startup that I've been helping a little bit and you can see like they did a pretty decent job based on the screenshot as well. Um, so yeah, there's lots of other tools also available for this. Yeah. Well, why don't we talk about this tool a little bit because like you must be crazy to enter this market right now, but like I think Magic Patterns has some unique features, right? You want Yeah. Exactly. So, one of the interesting things actually about this space and why I started doing a little bit of like advisory work is because all the tools are converging on like the same thing, but I didn't really feel like any of them served product teams that well. I feel like we're kind of like trying to get it to work for our use case, if that makes sense, rather than the tool being built for us. Um, and so there's a couple of like features that they've shipped recently that I thought helped with that problem. So, like one example of that is um yeah, they have like a feedback collection UI now. So, you can configure questions. And if I just like pop this up in a new link here, um, we should be able to see that. So we see like, you know, you can configure kind of like regular product feedbackesque questions, right? Uh, ratings or whatever you think uh, you want there. And then also another one which is kind of odd to me is you can require a password on the link, which seems like it should be the default for all of them, but it's not. Um, so what that means is like when you publish with Bolt or Lovable or Vzero to a public URL, um, it's a public URL, right? Like people can just access that link. So yeah, they built in this, you know, kind of more secure password protect feature here where uh you can do that. Yeah, I mean that that's like a no-brainer, right? Because a lot of times I'm prototyping something that I haven't launched yet. So like Yeah, you know. Yeah, agreed. I think it's honestly very strange, especially because it's I mean like I helped them kind of with this feature and it didn't it took them like a week to ship it. Yeah. So, um it's very strange to me that no one else has done this so far. And then yeah, they have some other kind of interesting things like probably one of the more interesting ones is the ability to put this on a canvas and then put different chats side by side. So this from like an organizational perspective is actually pretty useful because you can see like I've imported this design here and then if I want to test a different version of it, I can just fork it, get the copy and then hop into this one and have like, you know, the kind of the sideby-side view of like this is my new one here. I can just pop into it. Oh, really? Oh, like it just copies the code, too. Yeah, exactly. It copies the code. Um gets you like that side by side view, right? So, if you want to have more than one different um iteration of the same idea, you can just put them side by side. And this is actually something I don't know about you, but I've struggled with is like my both chat history is insane because I just have like a million different chats. Um so, keeping organized can be a bit of a struggle. That's actually really smart, man, because like yeah, when I use both of these tools, I always want to make a few variations and like uh I guess maybe I'm too dumb to like is there a way to because I always have to like revert to checkpoint or something to get the previous v variation. Yeah, it's a pain in the ass, you know. Yeah. I mean, the tools do support a similar workflow of like creating a fork. So, if I hop into here, I mean, I'll show you how it works in Bolt. It's super comp Well, maybe I'll do Vzero because it's easier. Uh so, in V 0, if you click this little fork button here, you get a copy, right? Okay. Um, the problem is though that like now I have these two copies in my chat history and that's kind of it in terms of how it's organized. Um, and so that's I think where there's some value in this type of approach here. Yeah. You want to see this stuff like all on the same screen, right? So you can Yeah. Yeah. Exactly. Um, so all right. Cool. So let's go back. So let's go back to our um Figma to prototype uh verdict. Which one is the best? Or may maybe magic pattern is the best since you're PMing it. Sure. Yeah. So, I think for the Figma import, I think Bolt still has it, honestly. Uh, it's still not perfect, but I think it's the closest. And then I think for like the screenshot reproduction, um, I mean, out of the

set, like maybe lovable, honestly, in terms of like the closest to the UI, but, uh, I think Magic Patterns did an all right job as well. Okay, got it. Yeah. Uh yeah, it's kind of funny because I remember lovable marketing a lot about like their Figma import, but it's kind of funny that the screenshot actually works better. Yeah. Yeah. For this use case at least. Yeah. One funny thing I've noticed actually about the marketing side of this is like these teams ship extremely quickly, but there's a ton of edge cases that I feel like don't get handled. And so the features like out there in the world, but it's not necessarily uh super great. And so I think like a good example of this actually is Bolt with the mobile apps. I don't know if you ever tried to build a mobile app with Bolt, but um in my experience, like I've tried it and more often than not, like maybe 50% of the time, it falls apart on me and I'm more technical than probably the average person. So, uh yeah, that one's a little bit tricky. Yeah. For mobile, I probably should use repet or something. Yeah. Yeah. I agree. But but uh yeah. So, okay, let's move on to the next use case um which is a JavaScript game. And I think you actually had a really awesome use case for this one. Yeah. Yeah. So, I tried to do a 2D Zeli game, which is a little ambitious, especially from a single prompt. As I mentioned, the only thing that I tried to fix was if there was an error, right? That was it. Um, but otherwise, I kind of just left it as it is. I actually got two functional games, which was surprising to me. Uh, so the tools that we used here were Firebase Studio, uh, we talked about a little bit previously, and this is the one you see on the screen here, and then I also used Replet, and then finally, Windsurf and Cursor. Okay. Yeah. Um, so let's show Firebase Studio first. Yeah. Yeah, this is Firebase Studio. Um, so there's nothing. I can't move. There's a red box over here. This is it. This is the whole thing. Um, so yeah, there's I can't do anything. And as I said, like I really wanted to try to make this fair. So I didn't try to fix anything. I just let it do its thing. I use the exact same prompt for all of them. So this was the result for Fire Race Studio. Uh, yeah. Not great. So need some improvements for Yeah, they do have one kind of like weird kind of cool feature though. So if you click on this button here, it opens it up in Excal the UI and you can draw on top of it and then use that as a prompt to go back in. It's a super cool idea, but also like I'd prefer that the you know the regular prompting works before they add stuff like this. Yeah, this looks like a bell somewhere. Yeah. Yeah, exactly. Yeah. So pretty neat, but um for now Yeah. I would rather just have regular, you know, prompting work. Got it. Yeah. Cool. Uh, from there we have wind surf. So with windsurf, I actually used the uh 4. 1 model, the new model from open AAI. And so this is our again single shot. It came out okay. Like I can attack. That's what I'm hitting space here. But I can't move. Also, the enemies don't move. U there's a tile set though. That's kind of cool. And uh yeah, I mean it's I could probably clean this up if I needed to. But again, I I wanted to leave it for the sake of fairness. Okay. And then uh next up, we have cursor. So I used sonnet for this one. And I'll just refresh the page here. So you can see that I have the ability to move. I have like this kind of pixelated tile set background. There is collision, right? So I can't like move through Well, except for there. I shouldn't be able to move through objects. Uh and it also randomizes the tile set, which is kind of neat. You can see here, right? Yeah. And there's like en enemies and there's enemies and there's an attack. Uh and a health bar for me and for them. I can't die. Oh wow. Uh but you can see my health bar kind of going down there at the top. So yeah, I mean it came out kind of reasonable to be honest. Uh for a single prompt and just to make sure your prompt is just like make 2D that was it. Yep. I didn't do anything else. Okay. Nope. Okay. Uh and then last but not least for sure. So we have Rocklet. Replit has like this kind of intro screen. It has a mute and unmute button. I couldn't figure out if there's actually any audio playing to be honest. I couldn't get it to work, but theoretically I guess it's there. Uh controls and so on. And then we have like this thing here. So this to me was actually incredibly impressive. Yeah, this is very impressive. Yeah. So we have um this enemy for some reason is locked in this kind of prison here, but we can go in and get him. And then we have again collision um with some objects here. We have like a tile set that is not randomized. and then like the panning and all this stuff. So, like I said, I was very pleasantly surprised with the result of this one. Yeah, this is very impressive, man. Like it also shows you where you're pointing and then and then you can attack or Yep. directional attacks. Yep. Okay. Wow. Yeah. So, yeah, I think Yeah. Go ahead. How long did it take for it to make this thing? Yeah, I think it was about 10 minutes or so. 7 to 10 minutes, something like that. So, not terrible. Well, that's I guess one difference with Replet, especially with the newer agent, is that each prompt will take like 7 to 10 minutes potentially in terms of the iteration. So, it's not a quick experience, but it can do a ton of work in one request. I have to give credit to representation. I think got a lot better like um it includes some of the best practices like it ask you for clarification, it takes screenshots, you know, it's like a lot better than before. Yeah, I think right now Replet's the only product that I would really consider to be like in a real agent. Obviously like cursor and you know the other ones have like an agent mode, but really the only agentic thing they're doing is deciding on what context to send and then potentially giving you commands to run, but mostly they're just editing files, right? Write writing code editing files. Replet will do like literally everything. It'll create a plan for you. It'll host it for you. It'll like it'll set up a database for you. Like you can do anything really kind of through the agent. Uh so yeah it's pretty cool. Yeah this is awesome. I mean like curses agent is just like you know yo hello just like do just like make all the changes to everything. Yeah. Exactly. Yeah. So they do also still have the assistant mode. They've kind of hidden it. They used to have it more uh visible in the UI but you can see now they've added this a warning that it's works best for users who are comfortable with you know basically more technical ideas. The assistant is very similar to the other tools. So it's like very small iterations where the agent is going to be like those massive iterations like building this whole game based on one prompt. So this actually kind of funny here. So like uh because I know all the people on Twitter you know are like v coding using basically cursor and maybe wind surf like I don't hear a lot of people building games with replet but based on this example replet is actually the best. Yeah I think replet is definitely up there. Also like the ability to host very easily is great too cuz I can just hit deploy and it's on the internet versus like with cursor you're going to sync it to GitHub and then figure out your hosting with like Netlefi or something else like it's there's a much higher barrier I think for average people to use cursor than there is to use replet. Yeah. And

also like uh like sometimes like the local and the remote can get out of sync like for this one you just have to like you just have one thing right. So yeah exactly. Okay. All right. So, so let let's get the word. This is kind of a controversial pick, but let's get the word to replet for building a game. Yep, I agree. Okay. And then you also have a personal app example, right? Yeah, exactly. So, um again, we used a couple different uh tools here. So, I used um you know, Firebase Studio, Replet, and then Cursen Windsurf once again. So, actually, maybe we'll start with Firebase Studio just for fun. So, here's what we got. I don't know what's going on over here. Um but we got the word goals. So No, hang on. What was your prompt to get? Yeah. So I actually gave it um like a more comprehensive prompt. Yeah. So I gave it user stories. I gave it the data model um and I gave it some mock data as well as um the kind of technology stack and so on. So and the components that I wanted like what the UI elements would be. So, I gave it more of like a proper starting point. Um, but yeah, we just got goals. Okay. Well, I guess the goal is to make this work. I mean, yeah, if you're the Google PM watching this, then yeah, best of luck to you. Yeah, I don't really know. And also, like it's very strange to me that they have this button here like edit the code. I feel like it like the average user doesn't want to click this button and hop into a different environment that's independent of the prompting environment where now I'm in a code editor. This is the idx product I was mentioning earlier. Okay. Um it's just strange, right? You have to switch back and forth in between them with this button. I mean that logo lo that logo look looks like a poo, man. I don't know what's going Yeah. It's just all very strange. So anyway, um yeah, I don't think Fire Studio is going to win this one. Okay. Uh here's Replet. So again, same spec. Replet again, it does a great job to be honest with like these types of tasks. So we have um by the way, context for this was it was like a task tracking app, but more focused on uh tracking your input rather than the output. So it's not goal focused. It's more like are you doing the things you're supposed to do? And so we have um like a calendar view. We have, you know, it tracks the when we've completed tasks. We have the different categories. I can log that I completed this. So I can say um I don't know whatever I want log that in there. Yep. Recent activity um you know create new categories and then like navigation as well for all these different things. Uh the stats like this is almost a complete application to be honest. There's very little like you could probably ship this as it is. So yeah super impressive. Wow. Okay. Yeah. Uh, and then here, I think this one was cursor. No, this one was winds surf. Sorry, because they're both local host, so I got to remember which one was which. Okay. Uh, this one was winds surf. So, again, same prompt. Uh, we have like some navbar stuff at the top, but it doesn't actually work. And then there's actually nowhere to enter anything. Uh, it just has some mock data, but nothing to really do here. Again, this was with GPT4. 1. Same prompt. Um, and there were actually a couple of errors that I had to kind of iterate through in order to get to this point from that initial prompt. Yeah, this one's weird, man. Like, what was this? Brush practice was that? Yeah. So, it tried to be clever or something. So, brain is like thinking work, brush is creative work, and book is learning. I don't know like teeth. I don't know about this one. Uh, and then yeah, the last one was cursor here. So, um, honestly, like pretty decent overall. We can log stuff. The UX or UI here is a little bit strange, right? Like the just the contrast on the page, but that's a pretty simple fix. Uh, and yeah, we can log that. There we have the filtering for the categories. And again, this is kind of reasonable. There's not much else going on here. Like, it's not a crazy complicated application, but it works. So, that's the UI looks a lot better than Way Surf. Yeah. Yeah, for sure. Okay. Yeah. Again, this has to do with the model as well. So like it's a combination of probably like Windserve as well as you know Sonnet 3. 7 versus GPT 4. 1. So this one was 3. 7 Sonnet and then this one was uh GPT 4. 1. Oh interesting because people are on Twitter are saying like uh 4. 1 is better now than Sonnet but I guess not quite on this example. I think it's hard to say like definitively what's best. A lot of it is personal preference as well, right? Um, but uh 4. 1 from what I've seen so far can do a better job of like some of the agentic stuff like tool calling or uh can be like you can give it negative prompts like don't do this thing and it actually doesn't do it rather than doing it because you've mentioned it. So there's some new capabilities I think that they've kind of built upon that are useful for developers, but I don't know in terms of like oneshot applications if 4. 1 is going to be the best. Yeah, you shouldn't expect to one shot stuff in cursor or wind surf. It's not designed to one shot. Yeah. Exactly. Yeah. In general, I guess that's another thing is like all these examples are all one shot, but usually what you end up doing is spending a fair amount of time iterating, right? Yeah. Um so this is a good kind of starting place if you're trying to figure out what tool you want to use uh to get that first result. That looks great, but beyond that, you might have different experiences with, you know, that iteration. Okay. But I think I like I mean I feel like we have to give the award to Replet again just based on this example. Yeah. Uh I think so honestly like Replet did definitely put together like the most complex application. The navigation works. Yeah. You know like it has this stats page. Again this is almost a functional app right and and it's saving uh it has a database. It's saving the stuff. Yep. Yeah. It has you know once again we have a client and a server. Um and then we have our database schema here. I didn't actually host the database, but again, it's a single click to set it up. That's really impressive because it's actually uh I don't know about you, but like it's really hard for me to get like a database and a server working in cursor. It's just like a pain in the ass. Yeah. Agreed. Yeah. I think again like Replet has found themselves in this position where like they're almost the best at the middle. Yeah. Which is building functional applications that are small in scope, right? you know, not like production professional engineering with an existing codebase. It's kind of this middle section. I just think honestly the demand for the middle section isn't as large as the demand for the other two sides. I think prototyping has a lot of demand and I think um you know like helping engineers has a lot of demand and then replet I think is great for like hobbyist people building their own apps. Yeah, that Yeah, that's a good point. I mean I guess it's pretty good prototyping too, but it's just like slower much slower than bolt because it has to build everything else. Yeah. Yeah, exactly. If you're waiting like 10 minutes per request for prototyping, like you're just going to give up, right? Okay. So, okay, cool. So, now let's let's go back to my slides. So, maybe let me share. Sure. Let me Sorry, I got to find it again. All right, there you go. Let me share my Okay. Um Okay. Okay. So now let let's actually pick the favorites. Uh hope hopefully we share the same favorites but I think from your examples it's pretty obvious. So I think for text prototype is probably both right. Yeah agreed. Okay both thing prototype based on our example is probably both too. Yep. Uh JavaScript game is probably replet. Oh Replet. Yep. Agreed. This is full stack personal app is probably replet too. Yep. Okay. Um, and we didn't talk about professional engineering, but um, I'm assuming it's either cursor or wer. I don't I guess I'm not really qualified to pick one. Yeah, I don't know either. I'm sorry to be honest. I think maybe we could do it based on this. Cursor is by far the most popular like by a large margin. Um, and I do think that they have some advantages just in like the UX of using the tool as compared to Windsurf like Yeah. And otherwise they've, as I've said, they mostly converge in terms of how they work. Yeah, I think it's like pretty neck and neck here. Uh and then but I think something big is happening with Windsor, right? There's like rumors that OpenAI will acquire them. Yeah. Yeah, for sure. I do think it kind of makes sense

because you know, one of the primary use cases that have come out of LMS is code generation. I mean, that's what we're all talking about, right? Whether it's prototyping or engineering or your own game or whatever, it's all codegen. And so for them not to have a distribution channel directly seems kind of strange to be honest. Uh and so yeah, I think it makes a ton of sense for them to do that. And I don't think the terminal thing is going to be the winner. Yeah. Okay. So this is like So we can tweet this out later, but this is kind of our current consensus and that that's so now let's talk about like um where this space is going, right? Like this space is like super hot, man. I thought AI models was super competitive, but this place is even more competitive, the AI models. Yeah. Yeah, agreed. Yeah, it's really interesting. So, I mean, we talked about uh one of the kind of like smaller players here uh that I'm working with, Magic Patterns, but there's actually like probably like 30 other ones. Like, so for every one of these big ones we've talked about, there's a bunch of other ones that are kind of trying to do the same thing. And I think the core problem is that we went from like category creation which is you all of a sudden you can use AI for coding to like where we are now where it's saturated in like what 5 months 6 months you can't even tell the difference between the anymore. And so I think the next stage here is actually just going to be that differentiation step right like how is VZero different from Lovable different from Bolt? and they're going to have to come up with some identity that they're targeting or persona of like who is the person who's going to use this tool because so far it's all been feature based. It's like come here to generate code. Yeah. Right. But now they can all do that. So yeah, curious what you think though. Um yeah, I think a couple things. So I I think your magic patterns demo is actually very good because I feel like a lot of these main players are not thinking about the whole user journey. They're just thinking about like generating code, right? like magic patterns. Maybe because of your advice, they're actually thinking about like, okay, if I'm going to do a prototyping tool, how do I get users to play with it? How to do the pass protection, you know, like they're thinking about the whole thing. Yeah, agreed. Yeah. I think I mean, it's actually funny. I started down the path of building my own, which we could talk about some other time because I was trying to figure out how to do this all by myself. It's not as easy as it seems. And then I kind of came to the realization I was like, I'm probably not going to compete in this space by myself. Um, so I had reached out to them and saw thought if you know I could kind of help someone build like the best PM tool then I'd get what I wanted which is the best PM tool. Um, so that's kind of the journey we're on right now. And so I so I do think like the I do think the prototyping thing is very different from the professional development use case. So, so I think there probably will be like one or two winners in the prototyping

space and like I I think um it could be one of the players we talk about or it could be like an established company like a Figma or like maybe Adobe or something coming space. I think one of the things we all kind of like underestimate is just how much enterprises value working with vendors they've already vetted. Yeah. Right. Like if you already bought Figma, even if Figma comes out with a worse product a year later, most enterprises are just going to pick Figma. Yeah. Right. And so I totally agree. Yeah. It's really hard on or Microsoft Microsoft's a classic example of this, right? Like Microsoft Teams, if you look at like the daily active users or weekly active users for Teams versus Slack, right? Like uh Teams blows Slack out of the water even though a lot of people think Teams is a worse product. Um it came out way later. Yeah. But somehow it still has like way more market share. because everyone already works with Microsoft, right? So, and and they just like bundle with all the other products. So, yeah, exactly. So, it's I mean it's kind of funny, but at the end of the day, it turns out distribution is the thing that matters. Uh once like the AI stuff kind of gets commoditized, right? uh yeah distriution and all the enterprise requirements and and like do we agree that prototype is very much in like I think it's the money I feel like is in the enterprise side you know like consumers prototyping maybe they'll do it like startups but like probably mostly on the enterprise side yeah I mean basically like I think the ideal place to be is uh where Figma is right so if you had the Figma for prototyping most of their money comes from enterprises like they have a free tier for you know I don't pay Figma I use Figma's product Uh, so yeah, I think by far like that's the place to be. It's just kind of hard right now. And also there's this other kind of looming threat which is what happens when OpenAI ships their own version of prototype like you know OpenAI is almost like the Apple in this space of like they could just eat your use case at some point and ship their own thing based on what's popular. Yeah. I mean given how popular so that that's probably like trend number two, right? Like given how much money is going into this co code cog genen stuff I feel like open AI and cloud and you know obviously goo Google are going to try to build their own thing or like acquire. Yeah. Agreed. Yeah. And then it'll be kind of a very similar layout or or landscape to many other products that we have nowadays, right? Like there's going to be the big players that are kind of established uh that most people use and then there'll be some specialized tools or use cases that smaller companies kind of take over. Uh but like on the professional development side like there probably is going to be a winner, right? Whatever this cursor wins or maybe you know open or something. Yeah. Because that's like um I don't know what IDs like most developers use that are not AI like I guess they use like just regular VS code. Yep. Yeah. So like Visual Studio Code or uh like you know Jet Brains is a company that just makes idees. Uh they have a bunch of popular different ideides depending on you the language you're programming in but um yeah it's I think honestly cursor is like far enough ahead that it's going to be hard for anyone to catch them. It's going to be interesting though if like uh open eyes buys windf right I guess if they buy windsfur it's kind of weird if winds offers like you can use cloud or some other model it's kind of weird. Yeah. They'll probably I mean I would guess they'll probably take it away eventually right? Yeah. Uh after like a year or something like that. But uh yeah, and like it's kind of like uh it's kind of crazy how badly uh GitHub copilot dropped the ball here because they used to be the most popular, right? Basically just took over. Yeah. It honestly it all started with the diffs. So like the view in in cursor where you can actually see the files that's changing. Yeah. In real time like the minus and plus because at the time GitHub copilot was just doing like more tab based changes. So like you'd be writing and it would recommend something. Oh okay. Like in the code itself. Yeah. But in cursor kind of I guess popularized the idea of having a chat in an IDE and then

you type in something and it shows you what's going to change and then from there we've gotten like more of the agentic stuff, right? Where it's like figuring out what to change running command terminals or terminal commands and all that stuff. Yeah. Like I almost never use the tab thing because I don't know how to write code. Yeah. So I always use the chat. Yeah. I think engineers I think are more on the other side to be honest. Uh cuz the chat is pretty dangerous. Like it's like here's 2500 lines of new code, figure out what happened and do you want it or not and it can take you just as long to go through that as it is to just do it yourself. Uh so you know Yeah. Yeah that makes sense. That makes sense. Yeah. So, so there's very clearly like uh there's like the, you know, not very technical people and then there's like actual engineers like two very different use cases and products. Yeah. Exactly. Yeah. It's actually somewhat interesting to me that cursive became popular in for like non-engineers. Yeah. And to me it almost shows you the amount of like pain people are willing to go through for the value and like there's a huge market there for nontechnical people in terms of just building a better product than an IDE. Because one other point in time is like anyone's been like I'm going to launch, you know, an IDE to do my job uh or to do a task I'm not usually doing like people are usually scared of IDs, right? Well, I mean again because like you know PMs have an inferiority called complex, man. Like we we've been trying to learn how to code for like you know a long time and like never made me you know finally we can actually start building random crap. Yeah. And basically just like oh yeah this is what it's like right like I can chip stuff. Yeah. That's probably a topic for another day, but um I've gone back and forth on the idea of like can regular people build production apps with these tools and like if you go on X or Twitter like the answer is like well obviously yes and you'd be dumb to not think so. Um but in my experience like I you know I've literally had hundreds of people go through my course so like start from zero and try to learn these tools as well as like myself as someone who's more technical and I think the answer is like pretty definitively no right now. Um because as soon as you get beyond like the bare minimum level of complexity, it the AI starts to fall apart and you really need to know what to do like yourself. And so unless you've basically unless you've done it before without AI or you've you know at least gone through the experience of learning how to ship applications, it's very hard to know how to fix stuff or what to do next. Yeah. You I guess you can for nontechnical people, you can build prototypes, you can build like very simple personal apps like the rep example we showed. Yeah. But yeah, to actually build like a proper app is very difficult. Like I my uh engineering friends, they're using cursor to vibe code too. And you know, if I be honest, like their apps are way better than mine. Yeah. You guys know what I'm talking about. Yeah. It turns out like all that technical knowledge is useful, right? Which is Yeah. A lot of it actually comes down to just prompting as well, which is like, you know, one good example um is if you want to have more than one page, like you have a nav bar, right, with more than one page. If you just say to the thing like set up routing or add routing, routing is the keyword that is a technical word that refers to the ability to switch between like navigation uh views. That's right. Yeah. Right. But most people don't know that. And so if you're saying like I want to have more pages that like that may or may not mean routing depending on how the AI interprets what you're saying. But if you say routing, it has a very distinct meaning that the AI understands. Yeah. Yeah, you need to understand all these like may maybe need to have a separate window open with like AI to translate your random requests into like technical terms. Yeah. And then Yeah. But yeah, that's a really good point. So, okay. So, TRD I don't think uh proper engineering is going to get this disrupted anytime soon. Yeah, agreed. Yeah. Y cool. Well, uh so Colin, where can people find you? This is an awesome discussion. Where can people find your course or start learning this stuff? Yeah. So, on Substack, I have a bunch of content on there. Uh so you can check that out if you want to kind of get started and then as mentioned teach a course on this topic. Uh so actually both technical foundations because I still believe that's important. So I teach that as well as AI prototyping over on Maven. So those are live cohort based courses about every month or so. Okay cool. All right man. Well it's always good to re with you man. Like it's always fun. So let's keep in touch and uh hopefully people learn from this video. Yeah. Awesome. Thanks Peter. All right.