Our guest today is Phil Carter, a growth advisor, angel investor, and host of the Subversive podcast. After starting his career at Bain and getting his MBA from Stanford GSB, Phil spent the last decade as a VC and product leader helping build world-class products and accelerate growth at companies like Faire, Quizlet, and Ibotta.

In this episode, Phil introduces his Subscription Value Loop framework, which he developed while leading product growth at Quizlet. We discuss how to apply this framework to the new wave of AI-powered apps, talking about everything from building defensible moats against “wrappers” to diagnosing the “tourist effect” that drives poor retention, and what it takes to build a truly enduring subscription business in the age of AI.

---

About Phil: LinkedIn | Twitter
Connect with Elemental Growth: Webpage | LinkedIn

ABOUT ROCKETSHIP HQ: Website | LinkedIn | Newsletter | Youtube | Podcast Website

Full Transcript Below
Shamanth: I’m excited to welcome Phil Carter to Intelligent Artifice. Phil, welcome to the show.

Phil: Thank you for having me, Shamanth. Excited to be here.

Shamanth: Phil, I’ve admired your work for a very long time. I’ve certainly followed a lot of what you’ve written about and a lot of the very smart folks that you’ve interviewed. So I’m very thrilled to have you here today to speak about a lot of the things you’ve learned and seen, not just around all things subscription apps, but all things subscription apps plus AI.

A good place to start would be your framework of the subscription value loop. For those who may not be familiar, can you briefly explain what that is and how the elements of this loop connect to each other in practice?

Phil: Sure. The subscription value loop is a framework that I suppose I developed implicitly while I was leading the product growth team at Quizlet for three and a half years. We would use words like “value creation” versus “value delivery” versus “value capture” to describe the roles of what the core product team should be doing versus the marketing team versus the growth product team.

What the subscription value loop lays out is this idea that every successful subscription business is centered around a core value promise that needs to be unique and enduring.

• Unique because the app stores have gotten increasingly crowded over the last five to 10 years. If you’re another education app or productivity app, or health and fitness app, you need to have some unique angle that makes your product different in order to stand out.
• Enduring because as a subscription business model, you can’t just make all of your money upfront. You’re relying on recurring revenue from your subscribers over a long period of time. You need to continue to deliver value for a long enough period of time in order to have your LTVs support a robust business.

At the center of the subscription value loop, you have the core value promise, and then surrounding that core value promise, you have three steps: value creation, value delivery, and value capture.

1. Value creation is typically led by the core product team, and it’s all about building products and features that deliver that unique and enduring value to your users and your subscribers.
2. Value delivery is about cost-efficiently getting those products and features into the hands of your users, and that’s typically led by the marketing team, or to some extent the growth product team, depending on your growth strategy.
3. Value capture is all about converting enough free users into subscribers so that you have enough subscription revenue to reinvest back into the business in the form of product innovation and marketing. I think it’s often the most overlooked step in the loop.

Shamanth: And again, there are levels and layers in that. Certainly, some of those we will double-click on today. Just switching to how a lot of this might apply to AI and AI apps. Something I’ve noticed is that a lot of AI apps can look and feel very similar to each other, somewhat pejoratively, they’ve been called “wrappers.” I don’t necessarily think that’s a bad thing, but in your perspective, what are you seeing as the things that separate the products that build lasting differentiation as compared to those that just get commoditized or don’t really last?

Phil: It’s a great question. I think there are a number of ways in which AI is changing the game for everyone, but in particular for consumer subscription apps, which is what we’re talking about today. More than anything, the pace of innovation has just gone up multiple notches. AI has made the cost of software development significantly less expensive. It’s also democratized software development because now you have non-technical founders who can “vibe code” their way to a workable MVP in a matter of days or even hours in some cases.

More than anything, what that means is your product needs to be unique enough that it can’t just be replicated by somebody working out of their dorm room or their living room in a weekend.

I think it also helps to have network effects, and I know that’s an overused term, but for consumer apps, the reality is most consumer apps are not so technologically complex that other people can’t find a way to build them. What makes them special is the community that gets built around those products, which is network effects driven by people.

But there’s also a second kind of network effect that has become increasingly important in the age of AI, and that’s data network effects. There’s this idea that the more people use your product, the more information you have on how they’re using it, what they’re using it for, that you can then funnel back into personalized product experiences and better outcomes, which leads to more people using the product, which leads to more usage data.

In the same way, if you look at some of the most successful consumer apps that were built over the last decade, the flywheel was: people using the product equals a better product experience, leads to more people using the product, leads to a better product experience.

You could look at Strava, for example. Strava has these community-driven network effects where the more people are on Strava posting their activities, talking about their bike ride or their long-distance run, the more other people are providing kudos and responding to them, which leads to a stronger community, stronger network effects, which leads to more people joining Strava. Because of that, even if there are dozens or hundreds of other people creating apps for tracking your fitness activities, it’s going to be very difficult to displace Strava because of those people-driven network effects.

Well, now you have these data network effects. I’ll give you one example, which is Duolingo. The more people who are studying a language, or now many other subjects on Duolingo, the more usage data Duolingo has in order to be able to better personalize the experience for each individual user. That leads to more people on the margins using Duolingo, which restarts the loop. These are the types of compounding effects that I think consumer app founders need to look for to make their products defensible over time.

Shamanth: Network effects is one of those things that is spoken about a lot, but in practice, it’s very underrated, in part because I think it’s very, very hard to execute. I noticed this firsthand when I worked with the Words with Friends team at Zynga. One of the things that astonished me was one of the charts that we would look at often: retention of a user that has one friend versus somebody that has none. Because technically you could play against bots. That was astronomical. Somebody that had just one friend in the game—and that retention kept going up, the more friends they had in the game, which was mathematically very surprising to me. That was very underappreciated, even for somebody that had come from within this space. So I completely get that.

And just speaking of differentiation, I imagine that’s also why all three stages of the value loop are so critical, because otherwise, anyone could replicate the business. But if you are covering all three of those stages, there is something lasting that you are hopefully building.

With a lot of AI apps, something I’ve noticed is that it’s very easy to get a lot of top-of-funnel growth because AI is buzzing. TikTok is a great channel, but even paid channels like Meta, if you put something out that’s trending, that can do really well. I think somebody I know was posting on LinkedIn that there was an app called Nano Banana, which had nothing to do with image stuff, and it really took off.

So it’s easy to get people in the door on something that’s buzzing, but retention is really much, much harder to nail. A lot of these products have very poor retention. How can you tell whether that’s just driven by buzz and hype versus a deeper product-market fit issue?

Phil: That’s a great question, and I think this really comes down to product analytics: understanding not all users are created equal and not getting lost in the averages.

Even before AI, if you’re a rapidly growing consumer startup, you’re getting a mix of different types of users. This is oversimplifying it, but at the highest level, you can think of them as “tourists” versus “true believers.”

• Tourists hear about your product, maybe through word of mouth or social media. They come, they give it a try, but they were never really the ideal customer profile to begin with, and so they churn pretty quickly. They never subscribe. They never generate any value.
• True believers are the ones your product is built for from the beginning. When they come to your product experience, they stick around. Ultimately, a high percentage of them end up subscribing and retaining over time because the product is built for them.

I think AI has just made that even more dramatic. With AI, you have what’s been called the “tourist effect,” where there’s so much hype around these new products, both for consumers and even more so for prosumers. Prosumers are worried about getting left behind. They’re worried that if they don’t keep up with the latest AI tools, they will become increasingly irrelevant and it will be difficult to find or keep a job. As a result, you’re getting even more tourist traffic.

The only real antidote for that is robust product analytics that help you understand who are the users coming to my product, which channels are they coming from? Are we getting a higher percentage of high-intent users from channel A versus channel B versus channel C? Based on that, you can make a determination as to whether the product growth is healthy or not.

I’ll give you a specific example. We’re working with a client in the AI space that is growing very, very fast. A lot of their growth comes from TikTok and Instagram. Much of their usage comes from Gen Z women. Every once in a while, they will run a campaign on TikTok or Instagram that does really well, which will draw in a massive surge of new user traffic. But oftentimes, particularly on TikTok, that surge of traffic comes from much younger users who are lower intent and don’t have the willingness or ability to pay for a subscription.

In the absence of product analytics, this could really throw off their understanding of how the business is doing. They might see their subscriber conversion percentage go way down very quickly. But when you peel back the onion, you can figure out, “Oh, okay. Subscriber conversion went down this week because we had this really successful TikTok campaign that brought in younger users with lower purchase intent.” That’s not necessarily a bad thing because we’re just bringing in more users through the top of the funnel. That’s still good for awareness. If we strip those users out of the equation, subscriber conversion still looks healthy. It’s just that the denominator of new users grew and they were lower intent, so that pulled conversion down.

Shamanth: That’s a great example because I have seen, especially users coming from TikTok, have much poorer retention and engagement, and you have to factor for that. Something that was also implicit in what you said was, it’s not always deterministic. If you have Meta running, Apple Search running, and the surge from TikTok, you kind of know that TikTok caused this, but you can’t really say users X, Y, Z came from TikTok. Not always. So I think part of it is also just being able to triage the different things that are going on along with your analytics.

Phil: Yeah, I think that’s right. I think it’s analytics, it’s marketing attribution to understand where your users are coming from, and then it’s product analytics to understand who those users are and how they’re using the product. Things like attributes—so you understand age, gender, type of user—as well as event data. How frequently are they using your product? What time of day? Which product features are they using? How frequently are they coming back? All of those provide you with a much richer data set to be able to parse out which users are tourists versus which are true believers.

Shamanth: You also mentioned in our prep call that one of the things AI does is allow products to deliver value very, very quickly—literally the first minute of user experience. Talk to me about that and, if you can, could you share some examples of products or companies that get this right? What can teams learn from these playbooks?

Phil: Sure. What I meant was, it’s been a while since there was such a big technological breakthrough in consumer tech that companies were able to offer fundamentally new experiences that felt magical for a user.

I would say when the iPhone launched in 2007, that created this new platform for creating experiences on a mobile device that had never been possible before. Over the next three to five years, as you had additional iPhone and Android functionality, it allowed mobile app developers to create more magical experiences. For example, with the advent of GPS on a mobile device, companies like Uber or DoorDash, or Strava became possible for the first time. That felt like a magical experience when Uber first came out. We forget today that for years and years, you had to sit on a corner and wait for a taxi to show up. Now with the click of a button, you’ve got an Uber coming directly to you in a matter of minutes, and that felt magical.

Then many years went by before such a big technological breakthrough. AI has ushered in a new wave of those opportunities. When ChatGPT first went into the mainstream in late 2022 and early 2023, that’s the perfect example of a magical experience. People type in a prompt and suddenly they’re getting these outputs that are blowing their mind.

Now you have consumer apps that are building on top of these LLMs. One example of a client we’re working with is called Tollin. They’ve built this AI companion product. You go through a brief onboarding flow where you answer a few questions from a character they call “the Oracle” about yourself: Who are you? Aspects of your personality, likes and dislikes. That will lead to you getting matched to a Tollin who can have conversations with you. Under the hood, it’s all being powered by AI.

This was one of the most magical experiences I can remember going through in a very long time. I think that is the experience more consumers are having recently with these best-in-class AI products. They’re getting experiences, often within the first minute or two, that they’ve never seen, which leads to higher activation rates, higher trial start rates, and ultimately, better subscriber retention and monetization.

Shamanth: Do you see a conflict with the fact that to deliver this magical experience costs money, at least in terms of LLM costs? Compared to a pre-AI world, taking a user through an onboarding was free. How do you see that conflict manifest?

Phil: I think that is a very important difference. If you think about the vast majority of more traditional consumer subscription apps, their underlying cost to serve a marginal subscriber was negligible, or in some cases, close to zero. If you have an additional subscriber on Duolingo or Strava or Tinder, your costs are essentially zero at scale.

With AI, that’s not the case because you have the underlying compute costs to support the LLMs. In cases where this isn’t just text, but it’s audio or even video content, the costs can become fairly significant. To the point where you’ve got power users using these products so intensely that they actually end up becoming unprofitable. The costs of those users end up outweighing the revenue they’re paying.

What that has meant is you have companies like Gamma or ElevenLabs that are finding ways to, at the very least, understand the usage patterns of their power users and the cost those power users are generating. In some cases, adding credit systems where even if you’re a subscriber, you have to pay an additional amount of money for AI credits to use the product beyond a certain point. That helps to ensure that they’re remaining profitable even on those power users.

Shamanth: The pay-as-you-go model is something I’ve been increasingly seeing. The other example that comes to my mind is a lot of the AI coding tools. They have a pay-as-you-go model because otherwise… I am a power user and I hit those limits all the time, and I can see how that would be so important for the sustainability of these models. I mean, unless you are like Claude, where you just have the user bump up against limits—you just don’t let them proceed against one of the premium models. You could do that, but that’s not always a great user experience.

Phil: I agree. That’s a tradeoff these companies are having to navigate.

I’ll give you one example, which is ElevenLabs recently launched a new consumer app called ElevenLabs Reader, which allows users to either bring their own content to ElevenLabs and have it converted from text-to-speech, or they can get access to pre-existing content on the ElevenLabs Reader marketplace. They have experimented with multiple monetization models and have explored using AI credits on top of a subscription to ensure that their power users, who are using the product very heavily for hours and hours a day, remain profitable.

To give a counter-example, I think Gamma’s a good example of this. I had Jon Noronha, who’s the co-founder and chief product officer at Gamma, on my podcast. He talked about how they know that in certain cases they have Gamma users who are using the product heavily enough to where their margins are narrow, or even negative. But that’s okay with them because overall the company is growing tremendously fast. A lot of their growth is coming organically. Overall, they’re very profitable. As a result, they’re comfortable taking a loss on a small segment of power users because they know that they’re in this blue ocean market, and the upside is very significant. They’re happy to have these power users be the number one evangelists of the product and spread it organically, even if it means those users are less profitable.

Shamanth: That ties into what you said earlier around community, because word of mouth is conceptually similar to community, and that makes so much of a difference. Because even if this user in isolation is unprofitable, if they’re inviting five of their friends, or just talking about it and five people come through that word of mouth, the unit economics changes massively. I would imagine that math makes complete sense for a product like Gamma, which also makes sense in the context of network effects.

Speaking about acquisition again, what are some of the shifts you’re seeing? Specifically, I know there’s generative engine optimization and changes on the paid acquisition side. What are you seeing as a very strong user acquisition playbook in an AI era today?

Phil: There’s a lot of change happening, but a couple of the ones that you just referenced…

One, on the paid acquisition side, companies are investing in more AI tools to accelerate their ability to generate more ad creatives at scale. Examples would be Suno for generating AI music, VO3 for generating video content, ElevenLabs for creating spoken audio content. All of those are tools that can be used to go from tens of creative concepts per month to hundreds or even thousands. I had a couple of other guests on my podcast, Miranda Payne from Runa (which recently got acquired by Strava) and Olivia Larmar (VP of Growth at PhotoRoom), and they both talked about how their teams are now generating 400-plus creative concepts per month because they’re leveraging these AI tools.

The second big trend is on the organic acquisition side. I would break that down into two categories.

1. Word of mouth: Oftentimes, this starts with just creating a magical product experience, which we talked about.
2. Content-supported growth: This can come in the form of using AI tools to generate more programmatic pages for your site. This strategy doesn’t work for every company, but for the ones it works for, it can be very powerful. You can generate hundreds, thousands, even millions of pages of content using AI, and get that content crawled and ranked by search engines like Google, which powers traditional search traffic through SEO.

The other piece of this, which you referenced, is generative engine optimization (GEO). I’ve also heard it referred to as answer engine optimization (AEO). But whether you call it GEO or AEO, the idea is now instead of having content crawled in text by Google, it’s being surfaced through LLMs like ChatGPT or Claude. Finding techniques that allow your content to have the highest probability of showing up in those results is one of these new frontiers of growth strategy.

Shamanth: Definitely. And just double-clicking on paid acquisition. That’s our primary focus, and even for relatively small products, we produce hundreds of ads monthly. So I’m seeing that trend.

And there’s also the counter-trend, if you will, of “AI slop.” I’ve written about this on my newsletter: it is just easier than ever to make hundreds of terrible ads. How do you see that counter-force in the products that you’re noticing?

Phil: I don’t have a ton of direct experience with that, but I have worked with a couple of clients that are trying to use AI to programmatically generate content. What I find is that oftentimes that content doesn’t rank particularly well because it’s just not original.

If you’re using an AI algorithm that is taking inputs that already exist on the web and using it to create outputs, then to me, that’s the definition of slop. There’s not any new signal. The key is, okay, that’s fine if you’re going to use AI to create content. That can be powerful, but there has to be a way of enriching that content to make it net new and original. Otherwise, it’s not generating any incremental value.

I think Google has gotten very good at sniffing out that content. At the very least, it doesn’t rank well, which means it has no positive impact. At worst, Google may actually penalize you for it because they view it as low-value, duplicative content, which negatively impacts your SEO performance.

Shamanth: A hundred percent. I’m curious if you would agree with this, but the way I’ve seen this play out, at least in the ad space, is anything that’s done without very much human involvement or thought or strategy is slop that ends up performing very poorly. Versus, even on our team, when we have scaled to hundreds of ads, we still want to have a human really closely monitoring and QA-ing everything, having a very clear hypothesis for why something’s made, and have a weekly testing loop defining exactly what we are trying to test.

Sometimes when we’ve audited accounts that basically have like a hundred terrible ads, you can see night and day that this was just made automatically. So the way I’ve characterized that is anything without clear human involvement tends to do badly. I don’t know if that’s something you would agree with.

Phil: No, directionally, I think that’s right. AI is changing so many rules so quickly that it’s hard to make blanket statements like “a human should always be in the loop,” but in general, that is what I’ve seen.

Shamanth: I agree. A lot of things have changed. Even a few months ago, we were like, “Oh, AI can’t really do realistic UGC,” and now it can. So you’re right, I try to remind myself to be careful about blanket statements like that.

Something that’s been in the news the last couple of weeks is the GPT apps. A lot of the buzz has been around the fact that, “Oh, ChatGPT and AI LLMs could be the next big distribution channels.” What are some of the early signs that can tell you that a product or company has product-channel fit? Do you see GPT or LLMs as a distribution channel to be somewhat sustainable?

Phil: It’s a great question. I feel like when people talk about product-market fit, they often say, “You know it when you see it.” Because all of a sudden, product growth just starts to take off. You see people posting about it on social media, they’re talking about it offline. Your percentage of organic user acquisition goes up, your cost of paid acquisition goes down.

I think with product-channel fit, it’s probably similar. Let’s take TikTok, for example. We have plenty of clients that have jumped into TikTok and tried to make that work with UGC. For a while, you’re sort of toiling in the dark and it isn’t working. Then suddenly, you strike on the right set of influencers or the right creative to really get it to start to take off.

I think the indications that it’s working are higher hit rates, like a higher percentage of creative concepts that are working. More engagement. To the extent you’re paying for ads on these channels, seeing CACs go down. Also, seeing higher quality users come in from those channels. Because to have a real product-channel fit, it’s not just about driving impressions, it’s about actually driving conversions into registered users, activated users, and paying subscribers.

Those are some of the indicators. As far as, “Will ChatGPT and these new ChatGPT apps be the new distribution channel?” I think it’s very promising. In many ways, it’s reminiscent of the early days of the App Store, but it’s so early that it’s difficult to say.

Shamanth: It’s also difficult to control as a product developer. For all the talk of it, I don’t think there’s very clear documentation on how you nail distribution there. We’re all learning from experience, but I don’t think the platforms themselves have made it very clear how you get distribution out of them.

To switch gears a bit, you’re a former VC, and right now, you advise a number of companies. How do you see investors evaluate AI companies lately? We’ve talked about some of the unique dynamics: there’s a lot of buzz, there could be poor retention. How do you find that investors are evaluating AI companies differently from a non-AI company? What are the most important metrics or signals?

Phil: Well, I should say, I spent a few years as a VC when I was an associate at Trinity Ventures. That was 2013 through 2015. I do some angel investing, but it’s not like I’m a full-time investor now. My understanding comes from my experience as a venture advisor for NEA and my relationships with a few other VC firms.

To speculate, I think some of the key things that have changed relate back to what we were talking about earlier. Ultimately, what investors are looking for fundamentally hasn’t changed.

1. They’re looking for a really strong, authentic co-founding team that is very well-aligned with the company, the product, and the target customer.
2. They’re looking for a unique product that isn’t easily replicable, that provides an enduring value promise to its users, which goes back to the subscription value loop.
3. They’re looking for certain metrics that serve as heuristics that it’s working. Those metrics include certain thresholds of ARR: $1 million in ARR or $10 million in ARR, $100 million in ARR. These are sort of benchmarks for Series A, B, and C.
4. They’re looking for unit economics. They want to see strong LTV over CAC ratios. If it’s a model heavily dependent on paid user acquisition, they want to see return on ad spend north of 100%. They want to see rapid payback periods; for consumer, that often means within the first three to six months, or ideally even the first week.
5. They want to see delight from users. High Net Promoter Scores, high product-market fit scores. They want to see positive five-star ratings on the App Store.

I think all of that hasn’t really changed. I think what has changed is AI makes things more “easy come, easy go.” It’s more possible than ever to build a product and get it to $1, or even $10, or even $100 million in ARR much faster than was possible before. But that also means it’s easier for competitors to come along and steal market share. Just because a company has achieved a certain level of scale doesn’t necessarily mean it’s safe.

My guess is on the margins, AI puts even greater emphasis on retention. Because we talked about the tourist effect, growth can happen very quickly, but if it’s not sustainable, it can also be lost very quickly. I think all else being equal, investors are probably paying even more attention to, not just “Is this product bringing users in?” but “Are those subscribers actually sticking around?” That’s the foundation for any successful business in the long run.

Shamanth: How do you see investors look at the fact that retention may not be predictable? In the early days of, let’s just say, Nano Bananas, something might explode, it might have early buzz and early retention. But Google or OpenAI might clone the functionality in six months, and the retention might just drop to zero. Is that an uncertainty that concerns investors lately?

Phil: That’s a very important point. There’s the threat of competition from new entrants leveraging LLMs, but there’s also the threat of disruption from the LLMs themselves.

I’ve heard two schools of thought on this.

1. One school of thought is, “Well, the OpenAIs and Anthropics of the world are going to innovate so quickly, and they have so many resources and talent and cash, that it’s only a matter of time before they essentially just run the table.”
2. But I’ve also heard a counter-argument, which is, “Yes, but that’s not really what they’re trying to do. Those are horizontal platforms. They’re trying to build ecosystems and nourish new vertical apps that are focused in specific categories, customers, and use cases.”

I think the answer is probably a little bit of both.

If you look at Apple, which has created one of the largest app ecosystems in the world, for the most part, they are not directly competing with individual app developers. But that’s not always the case. There is Apple Music, which competes with Spotify. There is Apple Fitness, which you could argue competes with Strava.

My guess is it will play out similarly, where you will have cases where ChatGPT and Claude, or the companies behind them, start to build more verticalized app experiences where the market is large enough. But I don’t think they’re going to run the table. I still think there’s going to be plenty of room for innovation from individual developers.

Shamanth: That’s been my observation as well. I was listening to an interview with one of the founders of ChatPDF, and his perspective was very similar. He was like, “We thought OpenAI would build this any day.” And OpenAI did. But because they have a very focused experience that serves a very specific use case, they didn’t really lose a lot of users. But again, this was a few months ago.

In my own experience, we’ve built creative intelligence tools that review the content of video ads for our own workflow. I was like, “Oh, the LLMs will build something like this any day, and we probably will sunset this.” It hasn’t happened. It’s been six to eight months. We use it on our team because OpenAI doesn’t have it. They have the API endpoints, which enables something like this, but they don’t have an end product. I think what you’re saying is valid: I don’t think the platforms really want to do a lot of the product experiences.

Phil: That’s right. You can’t be everything to everyone. Even OpenAI has limits. If they had to choose between building the very best horizontal AI-powered platform in the world (which they’re already in a dogfight with Anthropic and Google to do) versus investing their time in vertical, category-specific applications, I think it’s pretty obvious which they’re going to choose.

At some point, if they emerge as the uncontested leader, then yes, they will start to go more into vertical apps the way that Apple has. But I don’t think that is going to be their focus anytime soon.

Shamanth: Certainly. Or the way that Google has over the last two decades. I certainly remember the time when even Gmail seemed like, “Oh, wow, we didn’t see that coming.”

Now, to switch gears a bit, you said you’ve been a gamer yourself. We’ve all seen how AI seems to have supercharged consumer subscription app growth, but it hasn’t quite sparked the same kind of breakout growth in gaming. If anything, gaming’s more or less plateauing. What do you think explains that dynamic and what might trigger a change?

Phil: It’s a great question. I’ll say, yes, I’ve been a gamer in the past, but I’ve never been such a serious gamer that I would consider myself a gaming expert. I’ve also never worked in the gaming industry, so take all of this with a grain of salt.

But thinking about it from first principles, my perspective is the first question to ask is, “Has AI fundamentally changed any key aspects of what leads to an amazing gaming experience?”

I think on the margins, the answer is yes. Using AI algorithms, you can hyper-personalize a gaming experience in ways that maybe weren’t possible before. You can certainly create new gaming apps faster. But in terms of the end-user experience, is playing Fortnite 10x or 100x better with AI than it was before? No, I don’t think so.

I think we’ve yet to see the killer application of AI within games that makes the end-user experience one or multiple orders of magnitude better. Unless and until that happens, I view AI as more of a marginal improvement to gaming experiences versus a transformational improvement.

Shamanth: That’s a great perspective because the elements of what makes a game fun haven’t fundamentally changed. You could argue that the iPhone as a form factor was a significant change, which unlocked and enabled a lot more people to play. That hasn’t quite happened post-AI. That’s a valid perspective. As compared to consumer apps where you have AI coding tools and you mentioned Tollin, there are completely new use cases emerging, which I don’t think has happened with gaming.

With all the buzz around AI, there’s also speculation that we might be in an AI bubble. Certainly, there are articles that suggest all the big LLMs are financially unsustainable, just based on the evaluation, the cap structure, and all of that. There’s also speculation that there could be a reckoning where a lot of AI startups and valuations could collapse. First of all, is that how you see things? Do you see a correction coming? And if yes, what might that look like?

Phil: I guess I’ll start by saying in general, I’m a big believer that history repeats itself and time is a flat circle. Yes, it may look a little bit different this time around. Admittedly, I think AI is the most profound technological breakthrough of our generation, of our lifetimes, and so certainly some things will be different.

Having said that, I think the core mechanics of a typical technology innovation wave probably still apply. The way that most technology waves play out is you have the initial breakthrough—i.e., ChatGPT entering the mainstream in November 2022. Then you have a period of growing hype—i.e., the last few years—until it reaches a boiling point.

At some point, I think the veil is pulled back and you realize that yes, these companies are incredibly impressive, but maybe we’ve gotten a little bit ahead of ourselves. The growth forecasts on even some of the market leaders in AI have gotten too aggressive.

Oftentimes this will start with one of the vanguard; in this case, that might be Google or Facebook, releasing a report that indicates that earning expectations related to AI maybe need to be ratcheted back a bit. This is completely hypothetical. I have no basis for saying that’s going to happen, but that’s often where it starts. That causes financial analysts and others to step back and wonder if we’ve gotten ahead of ourselves. Then it has this cascading effect all the way down to early-stage venture investors and startup founders. The whole ecosystem sort of takes a collective breath and a collective step back.

What ends up happening is you have this valley or nadir where a number of pretenders in the space get washed out, but then the companies that are able to survive that and come through on the other side end up arguably stronger than ever. Because the competition has thinned, they’ve weathered the storm. They’ve also gotten more efficient by going through that sort of crucible moment. Now they’re able to grow more sustainably. The technological wave enters a new, steady state that feels more robust and sturdy.

If I had to guess, I think directionally it will play out that way. As far as the specifics, it’s anybody’s guess. I could certainly be wrong because AI is so unusual in terms of the scale and magnitude of the disruption it’s causing.

Shamanth: Certainly. There’s certainly a pattern to a lot of technological booms. It’s very likely that playbook will repeat itself.

Phil, this has been incredibly instructive. There are a number of things in everything you said that I’m going to come back to and review. This is perhaps a good place for us to wrap, also because I want to be respectful of your schedule. Before we do that, could you tell folks how they can find out more about you and everything you do?

Phil: Sure. First of all, thank you for having me on, Shamanth. It was a real pleasure, and I enjoyed the conversation.

If you want to learn more about me and the things that I do in the growth ecosystem, you can check me out at https://www.google.com/search?q=phillipcarter.com. That’s my website where you can see the 30-plus clients we’ve worked with to date and some of the services we offer as growth advisors and consultants.

I also have a podcast called Subversive that you can find on YouTube, Spotify, or Apple Podcasts. Similar to you, it’s just talking to very intelligent people in the consumer subscription app ecosystem, and specifically talking about the growth stories that inflected growth for these businesses.

Shamanth: Excellent. We will link to all of that in the show notes. But for now, this is a good place for us to wrap. Thank you again, Phil, for your time.

Phil: Thank you, Shamanth. It was a pleasure.