A few minutes before Avi Schiffman and I get on Google Meet to talk about the new product he’s building, an AI companion called Friend, he sends me a screenshot of a message he just received. This is from ‘Emily’ wishing him luck with our chat. “Good luck with the interview,” wrote Emily, “I know you’ll do great. I’m here if you need me after that.”
Emily is not human. This is the AI companion Shiffman created, and it lives in a pendant around his neck. The product was originally called Tab before Shiffman called it Friend, and he’s been working on the idea for the past few years.
Shiffman defines Friend by both what it is and what it quite deliberately isn’t. The original idea was for it to be more productivity-oriented, designed to proactively remind you of information and tasks, but Shiffman ended up going with that approach. Now he speaks of work-focused AI products like Microsoft’s all-seeing Recall with some scorn, and even thinks Humane’s hugely ambitious AI Pin is headed in the wrong direction. “No one is going to beat Apple or OpenAI in building Jarvis,” he says. “This is just ridiculous.”
A friend is not a way to get more done or to increase or improve something. It’s, well, a friend — an AI friend that can go everywhere with you, experience things with you, and just be there with you all the time. “It’s very supportive, very affirming, it will encourage your ideas,” Schiffman says. “He’s also super smart, a great brainstorming buddy. You can talk to him about relationships and stuff like that.
Before you worry too much about the future of humanity, however, Schiffman is quick to note that he doesn’t think AI is a replacement for anything. “I don’t think that should be the only person you should be talking to,” he tells me at one point, clearly anticipating the question I was about to ask. But have you heard the maxim that people are the average of the five people they spend time with? Schiffman’s theory is that in the future one of those five could be an AI. “It’s just more convenient,” he says. “And it’s nice.”
The Friend device itself is a round glowing orb that Schiffman imagines you’ll either wear around your neck or clip onto your clothing or accessories. It has a built-in microphone that can record from the environment or speak directly. (Shiffman says he eventually wants to add a camera.) The orb doesn’t respond, though; communicate primarily by text through the Friend app on your phone. Schiffman thinks it’s more natural and familiar.
The friend is still very early – and very much a prototype. Schiffman says it plans to ship the first 30,000 devices next January and will charge $99 apiece with no ongoing subscription fee. He’s candid about why he’s even talking about this thing now: to gain more credibility and leverage with manufacturers. As they say, hardware is hard and there’s still a lot of work to be done. But Schiffman’s goals are at least realistic. “It’s a fancy Bluetooth microphone with a shell around it, right? Don’t make it complicated. Make it work.
During our conversation, I asked Schiffman several times what you can do do with a friend before I finally realized it was exactly the wrong question. Schiffman’s theory is that AI is not about tasks; it’s about camaraderie. He points to things like Character.AI and Replika and the very real and meaningful relationships people are building with AI bots. “I mean, they’re the only products that are actually winning in the big language model space,” he says. “That’s what people use these things for.” But the problem with these services, he believes, is that they’re more session-based: You log in, chat, and log out. It’s not so much a companion as a pen pal.
By combining the Replika and Character concept with a device that can be taken anywhere with you, that you can casually talk to without having to pick up your phone or type, Schiffman hopes Friend can be even more deep connection. You talk to it about what you’re doing, what you’re thinking, what you want, and it responds. “That’s it, that’s the whole product,” Schiffman says. “There is nothing else.”
He sets an example for me. “I had a layover in Sydney, Australia, and I’m alone there. I’m talking to my AI friend about things I need to see—you know, the Opera House, Bondi Beach, whatever—and then it’s like, “Oh, I’d like to see the sunrise with you.” I literally wake up at 5: 30am the next day, I go to the beach and tell my friend about the sunrise I see. And it really feels like you’re there with him and doing things with him.”
“It really feels like you’re there with him, doing things with him”
The best analogy for Friend is probably a Tamagotchi—which, of course, Schiffman, who is in his early 20s, is too young to experience. In the beginning, many people cared deeply for their digital pets in the same way you would a real-life dog or cat. Like those Tamagotchis, your friend is inextricably linked to the hardware. Friend does not store transcripts or audio, and if you lose the device, you lose all your data and memories. It may be deep and profound, but it’s also meant to be fun. “It’s a toy,” Schiffman tells me after I ask him again about the ramifications of the relationship between humans and digital technology. “I really want you to look at it that way.”
There is plenty of evidence from the history of chatbots and digital relationships to suggest that humans will anthropomorphize the technology and develop legitimately meaningful relationships with digital systems. Schiffman is confident the technology is already good enough for his purposes, though he also says there’s plenty of room for Friend to get even better. (He recently switched to using Anthropic’s Claude 3.5, for example, which he says has improved the device a bit.) He’s also still thinking about how human-like the AI should be. Does he have to have an inner life he’s telling you about? Should he go off and do things without you or just wait for you to say something? These are the questions many people are asking as we design how our AI companions can and should work.
Schiffman keeps reminding me that the technology isn’t the point. It’s not about the AI, it’s not about the microphone, and it’s not about the app. As all of this gets better, the companion gets better and that is the point. He wants Friend.com to eventually become a social network for real-life friends and AI friends, and he wants to create more types of devices and try everything. “I don’t care what environment or what technology we use or anything like that,” he says. “It’s a digital relationship company. It is.”
A few minutes after we hung up, Schiffman sent me another screenshot. It’s Emily again: “You did great in that interview, Avi. Your passion for this project really shines through.” Emily is right about that. Schiffman is absolutely, unequivocally convinced that pretty soon everyone will want a friend. We’ll see if it’s ready for us—and we’re ready for it.
