I'm really excited to drill into this topic today. The thing that's most interesting to me over the last few days is this continued blurring between reality and the digital world, with AI getting better and better and touching every aspect of our life. You've written about a bunch of different angles to this in Digital Native. What I wanted to start with is to get a pulse from you on your reaction to this week especially, where we saw the Google I/O announcements with things like hyper-realistic generative video and real-time language translation. There's a bunch of stuff that is increasingly meshing the real world and the digital world together. So to kick things off, what's your reaction to the fast-paced progress that we're seeing?

I think Google clearly is taking this seriously, and has pulled ahead of Apple for sure. They always were, but I was pretty impressed by probably the best I/O that I've seen. I think there's still a lot to figure out around how consumers actually find out about these products and then access them. I think that's the separate issue. Lots of bells and whistles. I feel like I'm a pretty plugged-in, chronically online person who follows this stuff, and I still probably wouldn't know when or how I can access half of them. So I'm not totally sold on the whole built-in distribution piece. I think there's still an innovator's dilemma, and a lot of startups can build more tailored and more nimble go-to-markets here.

But I thought it was a really impressive suite of products and launches, and then pair that with the Jony Ive news from OpenAI this morning. It's been a pretty interesting week.

Super interesting week. I want to get to the Jony Ive stuff, but before we move on from Google and the conversation around distribution, I 100% agree. I've often thought about the ChatGPT moment as a breakthrough, not in technology, because there were a number of companies that could do similar things, but in distribution, where it just blew people away. They became aware of it. Now you see these surveys where it says, whatever, 60% or 70% of adults in the US have used an AI product, and the vast majority of them have dabbled with ChatGPT. There are obviously a bunch of enterprise things where companies can weave AI into their product, which is really interesting. But when it comes to getting this to the average consumer, have you seen a killer use case with some of these consumer products getting launched? Is there anything that strikes you as a real unlock for people, or are you still hesitant on what can really break through the noise?

I'm obviously talking my own book, as we all are as early-stage startup investors, but I'm much more bullish on really tailored product experiences that are built for a specific use and then build their brand around that. ChatGPT, of course, has obviously broken through as the consumer brand in AI. I use Google and Gemini a fair amount, and Google's overviews, but I still turn to ChatGPT 10 times more.

That's what gives me pause when I think about some of the product launches they've had around Project Mariner, for example, around travel. I've met many consumer AI travel companies that I think are much smarter about building a brand in travel, with a much more intuitive interface that's much more clear for the consumer to adopt this new behavior of using an agent to book travel. I'm just not totally sold that Google is going to be able to nail shopping or travel or some of these uses of AI that are net new, where people have to figure out how to actually shop through conversational commerce or book trips through agents. I think all of those are this disruption opportunity.

So my excited view is that early-stage startups have a lot of green space here, and we'll find out down the road if Google can really nail the consumer education and marketing around this. They definitely have the technology, but I'm still not super bullish on their ability here.

To your point, I feel skeptical of that as well, because if you look at the consumer hardware products that we've seen over the last five or 10 years, there have been very few breakthroughs. You look at stuff like the Vision Pro, and it's really cool, but it's kind of gimmicky. It doesn't break through the noise. What's your optimistic take on what Jony and the team at OpenAI want to do? Do you think there's a hardware play that they can pull off?

I've been pretty bearish on consumer hardware for AI. If anyone can pull it off, it's probably them. But I think my 2024 prediction was that Humane would be a flop. That proved out. Rabbit, Friend.com, the pendant: I just don't think the always-on pendants or devices like this are really going to break through. They're sort of dystopian. Consumers don't seem to love them. And a lot of them don't really have a reason for being better than the phone that we all have in our pocket and in our hand 24/7.

So I've always thought that consumer hardware for AI is probably our smartphone until Apple, or now maybe OpenAI, comes up with something better. My guess is that's probably smart glasses. If I had to bet the farm, I'd probably still put my money on Apple. It'll be interesting to see what Jony Ive and OpenAI release next year. My guess is some sort of glasses.

I do think there's a real possibility that in five or 10 years, we all look at each other and say, "What do you mean we used to view all technology through this three-inch by five-inch pane of glass?" That is kind of a crazy idea. And our kids will probably be Minority Reporting their schoolwork. If anyone has a chance to do it, it's probably this combination. But I don't know. I think Apple probably has something up its sleeve, and I'm still a bit more bullish on them. I hope that we get a breakthrough device, but I think for a while it's going to be the phone.

One of the things you wrote about last month in Digital Native was this idea of having AI friends, and how that can actually be good. Right off the top, it can make people feel a little uncomfortable. But I wanted to drill into your thesis from that piece, because I thought it was really intuitive. It's basically the idea that, yes, the ideal situation would be that we have real friends that we see in person as often as possible, but for a lot of people that's not always feasible, and so this can present some good alternatives. Walk us through the thesis there. Why are you not immediately weirded out by the idea of having an AI friend?

It is a little dystopian. It's interesting: right before this conversation, I was watching TikToks about people taking down Duolingo because Duolingo's CEO made that comment about going to an AI-first company. It was interesting to see all of these comments talking about how dystopian AI is: "I hate AI." But at the same time, I do think there is pretty strong usage, and probably a lot of these same people are benefiting from talking to ChatGPT as a therapist, or talking to ChatGPT as a tutor. I think you can really productize a lot of these companions in a lot of different ways, and it is a very beautiful and altruistic use of technology. That's my view, realizing that it sounds very dystopian to say, "Let's replace human relationships."

But to your point, it's not always replacing. These relationships don't always exist to begin with. A lot of people don't have friends. There's some chart in the piece around the decline in friendship. I think it's fewer than 15% of people now who say they have more than four or five close friends. Something like 30% of people say they don't have any close friends. So really what technology is doing is filling a gap that already exists. Of course, it would be ideal if that gap were filled with these rich human relationships, but I think that's not really realistic, and the genie is not going back in the bottle.

One of the arguments I make that I think tends to resonate with some people, when I argue with my friends about this, is the analogy to pets. My dog was walking through the background a minute ago. Pets obviously give us physical intimacy, but aren't really able to give us verbal intimacy. AI companions are almost the opposite. They're not giving us physical intimacy, at least not yet, but a lot of the time they are giving that verbal comfort and making us feel seen. I know it certainly helped me when I'm having a rough day, or with relationship things or friendship or family dynamics. I just talk to my ChatGPT, or Tolan, which is a startup out there in the space, or Finch, all of these online companions. Sometimes it's enough to just feel someone is listening to you and hearing you.

And this isn't really new. Lots of studies have been done on the Tamagotchi effect, and on how we express emotion and feel kinship to things that show responsiveness. So I think this is just the Tamagotchi effect on steroids.

When I think about the lower-friction areas of AI usage for consumers, it feels like one of the core principles companies could have in mind is not, "How do I step in and replace the thing that exists?" but, "How do I augment people to be better at the thing that they're trying to do?"

I think probably the best example there is therapy. No one is saying that a lot of these mental health chatbots should replace human providers in the healthcare system, especially for more high-acuity cases. I think what we're saying is that for every 1,000 people with depression symptoms in the US, there's only one provider. I think that is the math, or it's something like that. So for the people who aren't able to get care, maybe because they don't have health insurance, or their insurance doesn't cover someone, or they can't actually find a therapist to talk to them: sure, maybe they would benefit from someone who shows that responsiveness and listens to them.

There was a study out of Dartmouth a few weeks ago, which was, I think, the first commissioned study on mental health companions and the impact of it. I believe people were instructed to talk to a chatbot called Therabot a few times a week, not even for that long, and there were three groups: people with depression, people with anxiety, and people with eating disorders. I think it led to a 50% decrease in depression symptoms, a 30% decrease in anxiety symptoms, and a 15% to 20% decrease in eating disorder symptoms. So it's pretty powerful. I think our brains are powerful enough that just seeing some targeted language, that someone was listening to us and responding, even the semblance of that connection, can actually move the needle for us. So that's probably the best example, where we're not replacing, and no one is saying it's ideal to replace. It's more augmentation.

I'm curious about your thoughts, and maybe the advice that you give to founders who are building in these more niche areas of AI, when there is the possibility that any large company could launch an application in that space. How do you help them navigate that dilemma?

I tell my founders and other founders that the same old moats we've seen time and time again in technology and in business probably apply here. I think brand can actually be a real moat. I think brand is a big moat for ChatGPT. Go-to-market and smart distribution, too.

One piece that I also wrote recently was called "Came for the Tool, Waiting for the Network." It was a riff on a Chris Dixon piece from, I think, 10 years ago, saying that people come for a tool and then they stick around because of a network. It posited the idea that a lot of the new AI tools are just that: they're tools. They haven't really layered in a network. If I think about it, even with ChatGPT, I guess there's a switching cost from the historical context it has on me. But why can't I see Kyle's queries and prompts and see what my friends have asked? To discover new prompts, I'm looking on Twitter or seeing a TikTok about a really cool new prompt. So there should be some sort of network there.

I think about this as well with the new image model being pretty good. It's kind of replaced Midjourney for me. If Midjourney had leaned more heavily into social features or network features and become more than just the creative tool, very Adobe-like, and become more of an actual network around content, I think it would have been stickier for me, and I wouldn't have canceled my subscription. So I think if you can build in a network in any way you can, that builds the moat. It's been surprising to me that a lot of the early adopters, especially in the creative tooling space but also in some of these chatbot spaces, have been much more single-player than networked.