I Built Two Unicorns. Here’s The Only AI Startup I’d Build in 2026

AI doesn't really work. What do we do with that? Because people say, well, over time it'll get better. Like, all right, but you're building a company now. That's Jason Cohen. He's built $2 billion WP engine and smart Bear. He's been writing about startup strategy@asmartbear.com for almost 20 years. And he's one of the sharpest thinkers I know on what actually works in startups. I sat down with Jason to talk about AI. Not the hype, but the real question founders are dealing with right now. If he were starting a new company today, how would he think about it? He didn't give me a simple answer. He broke it into three types of AI products. Which one he'd bet on, which one he'd avoid, and why he thinks most founders are framing the whole thing wrong. Jason's been doing this for 25 years. Bootstrapped, venture backed, started companies, sold companies. So I asked him, if you were starting from scratch today, how would you approach AI? Yeah, it's a big topic. First of all, I separate how I use AI operationally to do stuff like write code or write marketing from AI that's in the product that the customers use, whether directly or indirectly. So first of all, take all that operational stuff, let's set that aside. Because I don't think that's what you were asking, but I find that in conversations sometimes people start confusing that it just makes it more difficult to deal with an already complicated or complex question. Right. So let's just ignore that. It is true that corporate budgets now are heavily biased toward things that are AI, whatever that means. And it's a fuzzy thing, like the companies themselves are not clear on what that means, so we have to be fuzzy. That tells me that whatever I do does need an AI component somehow because that's where the budgets are. It doesn't tell me what to build, but the idea of like, well, I just won't have AI at all. It'll just be a typical thing. That could be a good idea, by the way, but I would worry that, that I'll be fighting a budget battle and an attention battle. And so that doesn't feel like the easiest path. So if AI is where the budgets are, shouldn't we all just be building AI products? Here's where Jason sees founders going wrong. The wrong way to think of it is I need an AI product. That may be how the budget is. So you may, you know, that may be down the line, how you know, sometimes how you talk about it. But this is another thing I see people doing wrong constantly, which is thinking of AI as if people want AI as the problem they're solving. So let me put that differently. People have the same problems today as they've always had. Marketers want more leads. Sales wants to convert more leads to a sale that stay customer service wants to have good customer service and get good, you know, results from customers. You know, engineers want to write code that doesn't have bugs, that's maintained, manageable product managers want to build. Okay, everyone wants the same crap that they've always wanted. What you don't say is like, I need AI in sales. What you do say is, if I could 10x my outbound volume with the same conversion rate, that would sure be nice. So people talk about AI like it's part of the problem to solve or that people want AI False AI is part of the solution space. How is it that I can deliver more of what they already wanted because of AI? AI has made something possible that was previously impossible that they already wanted. So as soon as it's like, it's an AI voice thing, I'm like, I don't know what that means. Whereas if you said for restaurants, we take over their phone tree because we can do the menu, we can do the ordering, we can do hours, but we do it on the first ring and in 40 languages. Now, behind there is voice AI. That's what's behind that. Otherwise that's not possible. But the thing you're solving is your phone. Your phone calls are. Are now automated and awesome. And so I think when you stay focused on the problem that already existed, and AI is why you do something that was never done before or better. That's the right way to think of it. So I'd be thinking about AI as the solution space, not as people say. And they even say in their pitch decks as the problem space. Another thing I would do is I would say, look, AI doesn't really work. I know we. It's like. But when it does, it's amazing. Oh, I know. When it does is a pretty big qualifier. When I research stuff for the book, I would say at least half the time, it's simply wrong. Even when you do the deep summary, deep, you know, research, it sounds good when they say it. And then I go read the primary sources, and it's like, completely wrong half the time. And so what do we do with that? Because people say, well, over time it'll get better. Like, all right, but you're building company now. This is the tension Every founder is dealing with right now. You can't wait for AI to be perfect, but you also can't ignore that it's unreliable. I asked Jason how he thinks about what's actually buildable today. To me, there's three kind of categories of AI products right now. One is AI that the incumbents are inserting into existing products. There's like, notion, and you can talk to notion. And sheets. You can talk to sheets. That's not going very well. Like, it's not very useful. Right, okay, whatever. But of course they're doing that. What else are they going to do? I would do it too, but okay. The second kind is AI for experts. I'm already a software developer. Here's AI that helps me write code. I'm already a marketing writer. Here's AI that helps me write articles or do social media or something. I'm already a designer. Here's AI that helps me design. So AI for professionals. Bad news is that you're selling to only those professionals, not like the whole world or something. And you're up against the incumbents. Okay? But if as a startup, that's what you are, the good news is it's okay that the. The AI isn't perfect because it's an expert. So when the code is wrong, the expert can fix it. When the writing's bad or wrong, the marketer can fix it and so on. So it handles the fact that the AI is not perfect. This is why I like this category. The third category is AI for noobs or AI for Muggles. I like to say, right, I'm not a software engineer. I want to make an app. I'm not a writer, I want to write a book. I'm not a designer. I want a website. Now, that's good. I'm not saying, like, that's bad. I get it. Like, it's empowering. It's good. Like, I'm not against it by any means, but we're talking about what business I would build, right? And the problem here is you get 70%, 80%, and then you're stuck. And as a noob, you are actually stuck. When I vibe code the SaaS, and I don't know anything about code, and I'm like, 80%, I can't build a SaaS company. And I don't, you know, there's a lot of, like, I did this. Like, I know, but it's not a SaaS company, right? If that's what you were going for, that didn't happen. And you can go down the line, like if you're not this or that, it's not going to and you can't fix it, you can't take it down the line, you can't because you're stuck. So the good news is the market's bigger because there's, you know, a hundred times or a thousand times more people that want a website than people that can build a website. So hooray. But here's, here's the biggest problem is that the fact that AI doesn't really work is like a massive hindrance, possibly a 100% hindrance. So that's okay. Like if that's what you want to do, like take that, if you like that trade off, go for it, right? I'm not, I'm not, I'm not judging. I'm just saying what I think the trade offs are. For me personally, I'm always a problem solution, sort of a entrepreneur. Like everything I've done is like, oh, I can make this better, you know, whatever. Like there's this corporate thing that needs to be done or could be done. I can make it better. The typical B2B mindset. So that's what I have. AI doesn't work yet. So let's give it to people for whom that weakness is not a deal breaker. In a minute, Jason's going to talk about what he would do if he was building a new company using AI. But first, tiny seed applications for our spring 2026 accelerator batch are open now. Tiny Seed is the year long accelerator that I run for ambitious B2B SaaS, companies that are looking for the right amount of funding, a community of like minded founders and mentorship from roster of world class mentors. Jason is both an investor and a mentor in Tiny Seed and is kicking around in our private slack, occasionally dropping the kinds of insights you're hearing in this conversation. If you're building B2B SaaS and have at least 500 in MRR and you want founders like Jason in your corner, apply@tinyc.com apply. All right, back to Jason. So if I were doing a company, I would be solving a real problem that already existed. Of course that has budget, probably budget for AI. Okay, fine, I get it. I would use AI to make something possible that was previously impossible and I would do it for four experts so that the fact that AI didn't work well did not mean the product was useless or bad. Picking the right category is one thing, but Jason says there's another filter one he learned building WP Engine. You know WP engine from the beginning and even now we say like, oh, we make your site fast. And of course a fast website is good. Search engines rank it higher. People don't bounce off of it as much. There's data that shows E commerce sites that are fast, convert better media sites, get more hits, which means more money. Like perhaps it goes without saying, but I just said it, why fast sites are better, okay? And literally make more money money for people like e commerce and media. So, okay, that's, that's good. But if your site is 30% faster, is that enough to motivate someone to care, to search, to migrate their site to, you know, to not migrate away later? Like, is that good enough? I don't know. It's pretty weak. The reason we said four times faster is we had customer after customer where that they had literally data showing that. And anyway, when it's that much faster, you can just feel it. You can just see like, holy crap, what the hell's going on? If that's the reaction, that's all you have to know. Like, yep, that's. So when it's not just quote unquote better or faster, more efficient, cheaper, but five times better, three times cheaper, did it right where you could measure it, but like, you don't even need to to see like, holy crap. Like when it's that much, then something mundane and commodity like hosting and how fast it is, how secure is even a commodity thing. If it's a not 30% but 3, 5, 10x, it's no longer a commodity thing. That's now a substantially different thing. Busting you out, differentiating, earning a higher price and etc. So I say that in general, that's part of why we went in the commodity market and wpn. Just because we had a couple of things like speed and scale and, and service, which were like that. But it's also, I would bring back to the AI conversation when the AI helps me write. But like in the end of the day, like I'm, I'm still writing an article a day, just like a little bit faster. I don't know, that's something. But I'm not terribly compelled. But when I go from writing one article a week to two articles a day and they're good, okay, that's like dramatically changing what's going on. Now do I want, I personally don't want to do that. But like, okay, that's a product, right? Or I couldn't respond to this many, I don't know, posts on Twitter. What are you supposed to do in Social media, Right. But now with this, you can. Okay, so if it's in this like multiple of saving money, increasing time, having outcomes, now maybe that's really valuable and I'll be interested in that. So that's another thing I would expect from my AI. If AI is supposed to be so revolutionary, how come it's only increasing my performance by 20%? Right. It's just not worth the hassle and the wonder about where this is all going. So I would also be looking for something where the AI can really make me 10x. Like in coding, it's not. Right. Like, all the studies in real engineering departments is not that AI makes them 10 times better. The only people who claim that are the people people selling AI products, the users of it. You know, there might be individuals claiming that on Twitter because it's something to say, but all the studies show that's not true. What I find in coding, though, is there are certain places where it absolutely is ten or a hundred x. Like if I need to use a library I've never used before to do something, and it's like, oh, just use this. And it does. It just works. I'm like, okay, you definitely, you know, in. In 10 minutes, you definitely saved me a couple of days. Like, there's no doubt, right? Then there's other areas, like a big code base with a hundred developers where it's just like. I mean, it's just wrong so much. It's just. You're wrestling with it. It's actually kind of slower than just doing it. So to me, it's contextual. The question is, when is AI coding a big 10x er and when is it not is actually the question. So I'm saying that again because that's also part of my answer of what would I build? I would build something like that where, like, contextually I really can be like that. So something as broad as coding, that's too broad, and it's not Even true that AI in that broad of a context is a 10x er. So I try to find a product where it really is true that AI today, even, you know, with its fail, really is 3xing something. Hopefully 3xing more value for the customer, not just saving money. Saving money is a much weaker pitch. Hopefully it's 3xing the value, and it really does, and the weaknesses and the failings are fixed by the customer. Or we've picked a domain where AI really is much better. Not like general AI, I guess. So these are the kinds of things I would go through as I would tick through it. At this point, I noticed Jason hadn't mentioned competition once. Turned out that's intentional. My assumption is that every market of any reasonable size will be flooded with AI products. Maybe it already is. But my assumption is, if it isn't already, it's going to be. So, oh, well, what am I supposed to do about that? And what moat do I have? Like, nothing. Because we all have. We all use the same models. We're all making prompts like. And as you say, tech is usually not a moat. Okay. If you have very, very, very. You know, we can all pick out special cases where the tech is the moat. But that's the point. That's why it's usually not. So there's no moats there. Everyone's doing this stuff. So, like, what am I supposed. When the market is crowded with the same kind of. We all using the same tech, more or less. And again, I think, well, then I've just got to have such a great vision for my product, my ideal narrow customer, how I build for them, how good it is for that particular customer, because not everyone's going after that particular customer. So how can I just make an absolutely amazing product for that? Trusting in the thing we talked about earlier, that other people will also like that and want to join in. And so how can I find things like that adhere to those other criteria? There's a lot of Venn circles here, right? But of course there are, because we just said all these things have to go right for the company to work. So. Yeah, that's right. It's going to be a Venn diagram with lots of circles and a center that might not even be there be a center, which is kind of the point, but those are the kinds of things I would look at. Now, again, not trying to imply there's no other way to build a company. Right. Of course there will be successes that don't do what I just said, like. Like we've been saying. But in the spirit of that bootstrap machine where, you know, these are the things I would do that I think increase your chance of success or remove some kind of risk or at least lean into some kind of strength you have, or go with the grain of what's going on instead of against the grain, and therefore just like, hopefully make all these little probabilities be a little bit better than maybe they would have been. That's what I would do. And of course, no matter what I thought as customers actually used it, I would discover I was right about some things and wrong about some things and stuff I didn't think of, you know. So of course I would have to be following my nose after that. But this is what I would start with and then of course, be following my nose as soon as I could intersect it with real customers. I'm always struck by how deeply Jason has thought through pretty much any question I can throw at him. He's got a book coming out soon called Hidden Multipliers. I already pre ordered my copy and you should too. And if you haven't seen his microconft talk on this channel, it's our most viewed video of all time and you'll understand why. About two minutes in. Go watch it next. Thanks for watching. We'll see you next time.