#10: AI Companions (pt1)

Moving beyond the models.

I haven’t published here since Oct, 2021 so I wanted to give a little insight into what I’ve been up to…

I remember the last time I wrote here. I was sitting next to my friend Sydney in Venice at a coffee shop. I remember the moment very well. My fingers were feeling electric. There’s nothing more exciting than when your writing starts to take shape. I told her I was going to be posting every week! And I meant it, I swear.

A week went by, then a month, then a year. And here we are, nearly 4 years later.

Was I lying to her? No, I don’t think so. What’s more realistic is I was merely lying to myself. I know it sounds trivial, it’s just a blog after all. But why would I do that? Did I just get caught up in the emotion of the moment and not think it through? Of course you can always say “life happens” as a comforting excuse, or a lack of discipline, but I believe discipline should come naturally when you find something you deeply care about (as is the case with writing).

If you want to know the theme of my life over the last few years, it’s been that - stop lying to myself. I think we grow up learning how to lie to ourselves and we spend the rest of life learning how not to.

Anyways, in the spirit of not lying to myself, who knows when I will be posting here next, so I hope you get at least a little value out of this. (btw - I think my last post does a decent job on what this blog is about)

Everyone suddenly seems convinced that AI companions will be at the heart of the next major computing platform shift (Mustafa Suleyman saying that verbatim earlier this month). It all sounds a bit strange, no?? A bit silly. Even to me. After all, it's just a personality, a character that you’re interacting with. Can a character truly reshape how we use the internet?

If you've been here before, you know we've been working on avatars for quite a while. I've personally been thinking about them for a very long time. In 2017, we realized avatars could become a cool new way for people to represent and express themselves online. Avatars have since evolved into something much bigger and way more ambitious: a gateway for interacting directly with AI. What started out as visual self expression now feels like it could become the main way we access and use AI, and eventually, the internet as a whole.

It would be naive, even arrogant, to suggest there’s an “issue” with consumer AI today. ChatGPT has taken on a life of its own, bulldozing its way through society and and flattening opinions left and right. Everything is unfolding as the gods intended and it’s been incredibly fun and inspiring to watch. There is something I can speak to, though. Something I believe the industry is underestimating and if embraced, could unlock the next radical leap in the evolution of consumer AI.

First, ask anyone young what they use ChatGPT for and 9/10 will say for academic related tasks. Not even search (Google proving ChatGPT isn’t a threat quite yet). Very similar to what computers were first used for. That’s not because the model isn’t capable of more. It’s because that’s what the current interface invites them to do. It’s the same reason prompting as a skill isn’t a very realistic ask for Gen Z. Expecting them to “learn how to prompt” is like telling them to learn Photoshop just to make their Instagram pictures better - it’s not worth the time or effort and likely never will be. The whole reason the art-of-prompting even became a thing is because people don’t know how to intuitively use AI to its fullest extent yet. With the right interface or experience - a fun, personalized, revealing, interactive, and social one - prompt engineering wouldn’t be as necessary.

AI companions simply won’t flourish long term without the right interface. Text-based chatbots just won’t cut it, no matter how powerful the models behind them. OpenAI could be a $1tn company with the most advanced models but if I’m 18 years old just texting a bot that helps with homework, that’s really all it is then right? And if it didn’t help with my homework? I’m not confident I would use it that much. It doesn’t matter at that point how “real” it sounds, how funny it tries to be, or how well it knows my lunch. The experience - the interface - determines whether an AI companion becomes a genuine presence in someone’s life over time or just another tool in the bag.

Think back to Apple - 1984, 1998, and 2001. The original Mac + GUI, company saving G3, and beloved iPod. These weren’t just technical breakthroughs. They were interface revolutions. “1,000 songs in your pocket” was a strong selling point but that wasn’t really why people bought iPods. It was the interface that made that slogan real and the experience enjoyable. That damn wheel, I’m obsessed with that thing. Who isn’t! Jobs didn’t just build great products - he built great experiences wrapped in an interface so intuitive and so profound that the Buddha himself might have crafted it. The technology just followed.

You could say consumer AI is still in its 'pre-Apple' era - a generic phase where everything looks and feels the same. And honestly? It’s all kind of boring. I don’t necessarily think there’s anything wrong with that yet. 400 million other people don’t think there’s anything wrong with that either. That said, I also buy my lamps from Target. Generic as hell lamps. I love them… until I have to move or I find something better. Then I leave them on the sidewalk.

Generic is fine. It’s simple, it’s easy, it works. But you know what generic doesn’t do? Build intimate and long lasting relationships. When my needs change, my products change. You don’t build a relationship with generic. You use it. You extract value from it, and then you move on to shinier things. A generic interface leads to a generic, hollow, and easily replaceable relationship. Generic leaves room for quality to take its place.

This might be a bold statement, but I believe it’s true: if one of the other models (Claude, etc) launched an aggressive Gen Z marketing campaign - Charli XCX using it at Coachella - they would switch! I swear. As long as it still helped them with school, they’d make the jump. Honestly even if it didn’t help with school, they would at least add it to their arsenal. The current moat isn’t loyalty or value. It’s habit. Habits can be replaced. Habits can be influenced. Habits can be left on the sidewalk.

OpenAI has indeed leaned into personalization and memory. Those systems still need user input, though. How much true personalization can be achieved when the product so far has revolved around finishing biology homework 50 times? How long will it take for a given user to provide enough input to truly reap the real benefits of LLM memory? Early adopters will disagree, but you are the 1%.

The thing is AI has already changed so many lives - including mine - in incredibly deep and personal ways, but getting to that point took constant experimentation, trial and error, and a whole lot of time. That path isn’t realistic for young consumers. The solution isn’t teaching them more - it’s removing the need to teach them at all. What will matter is not model quality, but rather the interface and experience which unlocks and surfaces the most valuable use cases. If model A is 50% lower quality but I know how to use it in 10 meaningful ways, while model B is technically better but only useful to me in 1, then model A wins. Whatever model is attached to that winning interface will reap the benefits.

The next era of consumer AI won’t be defined by more creativity or boosts in accuracy. It will be about the interface. It will be about personalization. It will be about making AI feel less like a tutor or a “knock off human” and more like something - or someone - that truly understands you, plays with you, and guides you. It will enable you to have conversations you never thought you’d have, learn things you never thought you’d learn, laugh as if it was your best friend, and discover sides of yourself you didn’t know were there. AI can help young people develop a stronger sense of identity, a stronger sense of independent thought. It can help them develop a passion for curiosity. It can help them take meaningful action in their lives.

TikTok has proven to be a mixed bag. On one hand, it’s given birth to an entire generation of new entrepreneurs, but on the other it’s turning the youth into zombies of comparison. I used to be a huge proponent, but now it kind of sucks to see how it’s played out. We make and consume content like our neighbors and eventually start thinking like them too. We offload our sense of self into an abyss of distraction. We’re getting lazy with ourselves and we’re losing touch. The algorithms so far are not personal, they just disguise themselves as such. AI can be extremely personal, though. It can truly counteract the algorithmic effect when not tied to one of them itself. It can push us inward instead of outward. It can help us discover what makes us, us. That’s when AI stops being just a tool and starts becoming a true companion.