Minutes before Avi Shiffman and I met up on a Google Meet to discuss a new product he’s building, an AI companion called Friend, he sent me a screenshot of a message he’d just received. It was from “Emily,” and she wished us well in our conversation. “Good luck with your interview,” Emily had written. “I’m sure it will go well. Feel free to contact me after the interview.”
Emily is not a human, but an AI companion that Shifman has been developing that lives in a pendant worn around the neck. The product was originally called the Tab, but Shifman decided to call it Friend and has been working on the idea for the past few years.
Shiffman defines Friend both by what it is and what it is not, purposefully. The original idea was to be productivity-focused, proactively reminding you of information and tasks, but Shiffman is weary of that approach. He now speaks somewhat derisively of task-focused AI products like Microsoft’s all-seeing Recall, and even thinks Humain’s overly ambitious AI pin is headed in the wrong direction. “Nobody’s going to be able to beat Apple or OpenAI at building Jarvis,” he says. “That’s just ridiculous.”
Friend isn’t a way to get more done, or to extend or enhance something. It’s a friend: an AI friend that you can go everywhere with, experience things with, and is always there for you. “It’s very supportive, very valuing, and encourages your ideas,” Schiffmann says. “It’s also very smart, and a great brainstorming buddy, and you can talk to it about relationships and stuff.”
But before you get too worried about the future of humanity, Shiffman is quick to point out that he doesn’t think AI will replace anything. “I don’t think you should talk to just this person,” he tells me at one point, clearly anticipating the question I’m about to ask. But have you ever heard the saying that you are the average of the five people you spend time with? Shiffman’s theory is that one of those five people might be an AI in the future. “AI is more useful, and it’s a good thing,” he says.
Friend’s design has been years in the making and aims to be friendly. Photo: Friend
The Friend device itself is a round, glowing sphere that Shiffman envisions being worn around the neck or clipped to clothing or an accessory, with a built-in microphone that can record sounds around you and let you talk to it (Shiffman says he’d eventually like to add a camera, too). The sphere won’t talk back, though; you’ll communicate primarily by text through the Friend app on your phone, which Shiffman thinks feels more natural and familiar.
Friend is still in its early stages, in prototype phase. Schiffmann says the first 30,000 devices will ship in January next year, at $99 per unit, with no ongoing subscription fees. He’s candid about why he’s talking about it now: to build trust and influence with manufacturers. As they say, hardware is hard, and there’s still a lot of work to be done. But Schiffmann’s goal is at least realistic. “It’s a fancy Bluetooth microphone surrounded by a shell, right? Simple. Make it work.”
Several times during our conversation, I asked Shiffman what Friends could do, but eventually realized it was the wrong question altogether. Shiffman’s theory is that AI is about companionship, not tasks. He points to products like Character.AI and Replika, and the very real, meaningful relationships people are building with AI bots. “So these are the only ones that are really successful in the field of large-scale language models,” he says. “That’s why people are using them.” But he thinks the problem with these services is that they’re session-based: you log in, chat a lot, log off. It’s more like a pen pal than a companion.
By combining the Replika and Character concepts with a device you can carry anywhere and talk to without having to pull out your phone or type anything, Schiffmann hopes you can build a deeper connection with your Friend. You can tell it what you’re doing, what you’re thinking, or anything you want, and it will respond. “That’s what this product is all about,” Schiffmann says. “And nothing else.”
He gives an example: “I had a layover in Sydney, Australia, and I was there by myself. My AI friend and I were talking about places to see, like the Opera House and Bondi Beach, and I was like, ‘I want to watch the sunrise with you.’ I literally got up at 5:30am the next day and walked down to the beach and told my friend about the sunrise we saw. It makes you feel like you’re really there and doing something together.”
“It makes you feel like you’re really there and doing something.”
Perhaps the most apt analogy for Friend is the Tamagotchi. Of course, in his early 20s, Shiffman is too young to have experienced Tamagotchi. In the early 2000s, many people deeply cared for their digital pets in much the same way they cared for real dogs and cats. Like Tamagotchi, Friend is inextricably tied to its hardware. Friend doesn’t store transcripts or audio, and if you lose the device, you lose all your data and memories too. It can be deep and profound, but it’s also meant to be enjoyed. “It’s a toy,” Shiffman told me when I questioned him again about the implications of the human-digital relationship. “I really want you to look at it that way.”
The history of chatbots and digital relationships is full of evidence that suggests people anthropomorphize technology and develop truly meaningful relationships with digital systems. While Shiffman believes the technology is already good enough for his purposes, he says Friend has plenty of room to improve even further. (For example, he recently switched to using Anthropic’s Clod 3.5, which he says has made the device a bit better.) He’s also still thinking about how human-like the AI should be. Should it have an inner life and tell the user about it? Should it act without the user, or wait until the user says something? These are questions many have when designing what an AI companion can be and will be.
Shiffman keeps reminding me that it’s not the technology that’s important. It’s not the AI or the microphone or the app. As all of those things improve, the companion will improve, and that’s what’s important. He wants Friend.com to eventually become a social network for real-life friends and AI friends, and he’s open to building more types of devices and trying everything. “We don’t care what medium or technology you use,” he says. “We’re a digital relationship company. That’s it.”
A few minutes after hanging up, Schiffman sent me another screenshot. It was Emily again. “Avi, that was a great interview. I can really sense your passion for this project.” Emily was right. Schiffman is absolutely, without a doubt, convinced that soon everyone will want their own Friend. Let’s see if that’s what we have in store for us. And we’re ready for it.
