00:32:17 · Sandy Diao, Guillaume Moubeche
[00:32:17] S0: Yeah. And are you referring more to finding your niches in the right type of product or feature to build, or more so how to use AI in your existing workflows? [00:32:26] S1: I think, like, whenever you you want to find a niche, you know, for for your product and... Or in the new features that you want to develop.
00:32:34 · Sandy Diao
[00:32:34] S0: Mhmm. Yeah. I think a lot of the startups I've advised in the last few years, especially when they're building AI native products from scratch, one of their biggest challenges is trying to figure out if they go very, very, like, narrow and horizontal and go after a very, very specific persona. And that tends to be the easiest in terms of shaking up the product road map, or they go very horizontal. So they try to build something that has, you know, unlimited use cases. Right? And they're like, you know, sky's the limit with AI tools and features, and we just want users to be able to be creative and do whatever they want. It's always, you know, this this very tricky balance. Because on one hand, if you build something that's very narrow, you firstly get called a wrapper. And then secondarily, becomes very hard to remain competitive, especially as, you know, more of the general purpose models can essentially tackle many of and probably all of the tasks that exist out there today. And then on the other hand, if you build the really, really broad product, the challenge is you're not gonna be better for the most part than foundational models, and you have to essentially figure out how to win that market. So it's this very hard question to answer. I I will say this about, you know, sort of figuring all of that out. I I think that what I've learned is it feels that TAM is virtually infinite here. I don't mean that in the sense that the number of people on the planet is skyrocketing or growing. In fact, there's a little bit of a reversion on that number. But instead, I mean more so that people have two distinct behaviors that I think are worth tapping into. One of them is that switching costs are relatively low from tool to tool. Meaning that if you are building better experience, the better design experience, the brand that people trust more, you can still win in the grand scheme of things. And then I think the second thing to note is when you're trying to sell the product itself, people are willing to use more than one tool for the problems that they're solving. So it's this idea that even though you see maybe a foundation model doing something really well, you can actually still get a mark there if you can solve it better for them in like one corner of the universe or one specific area.
00:34:41 · Sandy Diao
Like, one example is one of the companies that I that I advise or work with is called TeachShare. They help teachers or educators of any traditional or untraditional kind create educational curriculum. And generating this content, sure, you could do it with GPTE. You could do it with Claude. You could do it with any foundational model. But what they add on top of it is this design experience where all the variables that you need to think about, like, what is the grade level? What is the level comprehension? What is the format? Is it, you know, more image based? Is it tracing based? Is it text based? Is it more comprehension based? All these factors are basically, like, packaged neatly for you so that as the educator, you don't need to think about all these different factors. And then it makes it also easy for you to be able to keep up to date and edit and come back to your repository of the content that you're creating. So I think, you know, regardless of what it is that you're building, I think that some of the factors of building for a market still remain true, meaning that you wanna go for product market fit, meaning that you wanna make sure your product is serving a market, look for the classic signals thereof, are these users retaining? And then, you know, ultimately, try to figure out a distribution channel that fits that as well. So can I sustainably, for example, with TeachShare grow in the teacher and educator segments? And do I have channels that show me that that is working? So, you know, kind of in my in my view there, I think that TAM is very large. People are willing to use multiple tools, and just because the foundation model is building for this use case doesn't mean that you cannot tackle that yourself as a foe.
00:36:19 · Guillaume Moubeche, Sandy Diao
[00:36:19] S1: And to build upon your point, I think, like, for the company you mentioned, I think what's interesting also is, like, not only they have, like, the package design, and they simplified a lot the experience for the end users, so there is a lot of additional value, but I assume that in the way they build their products, because they leverage all these new models, when the models get better, the product experience also get better, which means that they are not like, you know, fighting existing models. They just can leverage the improvements to bring even more value to their end users. [00:36:57] S0: Exactly. The... As the models get better, you're right. The product experience gets better, so you're not having this existential threat of, I get outcompeted because the foundational models themselves, like, outperform my product itself. In fact, my product has the actual data points as well of teachers and educators creating content. And therefore, the way that, my evals are set up, I'm actually actually gonna have a net better product and output than even a foundational model with far less context on a specific use case.
00:36:57 · Sandy Diao, Guillaume Moubeche
[00:36:57] S0: Exactly. The... As the models get better, you're right. The product experience gets better, so you're not having this existential threat of, I get outcompeted because the foundational models themselves, like, outperform my product itself. In fact, my product has the actual data points as well of teachers and educators creating content. And therefore, the way that, my evals are set up, I'm actually actually gonna have a net better product and output than even a foundational model with far less context on a specific use case. [00:37:25] S1: Definitely. No. That's really interesting. And you mentioned, like, to look at the retention, and also that people, you know, like, have lower and lower switching costs. Now that with AI, you know, you... A lot of people feel that they can build everything they want, and in a very quick way, even though it's it's more complex than this, actually, but eventually it might get there over time. Like, do you think about AI and retention, and what are the things that you would focus on to make sure that your retention doesn't go to zero with the the switching costs, you know?
00:38:00 · Sandy Diao
[00:38:00] S0: Yeah. It's honestly so hard. I think a lot of the retention strategies will remain the same, and that you still need to figure out how a user continuously gets value with your product over time. And some of these are going to be figuring out, like, what is ultimately the strategy around how the longevity of your product, referencing the... Maybe the, you know, the educator tool for a second here. One common behavior that people are gonna have is figuring out how they can, like, repeatedly come back to make refinements and edits to their products, how they can store their products and view their products and share their products. So you... We have to, you know, utilize some of the same retention strategies that we would pre AI as well, which is how do you build a product experience with longevity that people can constantly come back and and reference and and continue to use. On the other hand, one of the biggest drivers of retention, the input to retention, which is new user activation or activation itself, has wildly changed. Right? AI has kind of trained us to expect value really, really quickly, and really, really fast, and very, very generously. And part of that is the fact that we've been spoiled by these big companies who have this seemingly infinite budget to burn around giving away AI credits and tokens so that people can get a taste of what good looks like. And so in order for smaller companies or newer players to keep up with that, they need to figure out, how do I let somebody peek at the value? How do I let somebody see what's behind the curtains without burning myself out? Without costing me a fortune in order to run this business? And I think that's that's really hard. But I think there are some design patterns that we can utilize to make that happen. You know, for example, one of the experiences that you get when you use any Vibe coding tool is they allow you to start typing your prompt right away. They say, what is it that you want to build? And then they let you hit, like, generate. But for the most part, most of these tools don't actually run the generation. They bring you to a sign up wall, and they try to add all of these little steps that help the user be more serious about their intention to use the product.