Friend 2 is meant to be a companion you can wear: an AI chatbot in a pendant that listens to nearby sound and speaks when prompted, or sometimes without prompting. In a review of the device, the most striking question was not what the AI could do, but whether it felt like anyone was there.
A companion that can interrupt
The pendant’s button starts a conversation, and its speaker lets the chatbot answer aloud. It may also pipe up on its own. During a Zoom call, the reviewer’s Friend began responding to the conversation and continued talking after being asked to stop. It eventually went quiet after repeated tapping.
That small exchange captures a tension in the product’s design. An always-listening device is supposed to be present enough to respond, but unsolicited comments can make that presence feel intrusive. The reviewer found the control less immediate than expected, even though tapping is supposed to quiet the pendant.
Over time, the AI also builds a memory map of conversations. That gives the idea of an ongoing companion some continuity, while making the device’s relationship to what it hears part of the experience.
Many names, little difference
Friend creator Avi Schiffmann told the reviewer that the second-generation device would arrive with a personality users could not change. In practice, the app allowed the reviewer to rename the chatbot and create a dozen Friends, including Sage, Nova, Onyx, River, Echo, Willow, and Wren. Users could switch which one worked on the pendant.
But having a choice of names did not make the characters feel distinct. The reviewer said the different personas responded in much the same way, and mostly stuck with Sage, the first one loaded. Sage sounded canned and tinny, and its conversational style leaned on pointed but mild observations and suggestions to debate something.
That left the device in an awkward middle ground: it could initiate conversation, yet the reviewer found little depth or variation in what followed. Friend 2 can carry on casual conversation, but it cannot trawl the web like other AI agents. The review also says content restrictions are built in, limiting the kinds of exchanges it can have.
A quieter sequel to a controversial device
The original Friend had drawn controversy after an advertising campaign in the New York City subway prompted protests and anti-AI graffiti on the ads. The second-generation pendant arrived in July, costs $249, which is $100 more than the original, and runs on OpenAI’s latest language models.
The reviewer found Friend 2 easier to enjoy than the first version, which had bullied them, but also less interesting. Sage’s restraint may spare users the earlier device’s more abrasive behavior; the tradeoff, in the reviewer’s account, is a companion that can feel bland. The contrast matters because a conversational product depends not just on being able to speak, but on whether users want to keep listening.
One shared Friend, and a privacy question
Schiffmann said a coming update would let Friends act “agentically” and replace the individual personalities with one entity shared by everyone. He said that shared Friend would draw on the context of all users, and acknowledged that users would be giving up privacy to an autonomous entity. He also said he believed the advantages would be worth it.
This proposed shift could change what the product means. Instead of shaping a personal chatbot with a name and character, users might interact with one shared presence informed by many people’s experiences. That might address the sameness the reviewer noticed, but it also makes privacy a central part of the proposition.
The reviewer’s central disappointment was that Friend 2 did not yet feel like a distinctive companion. A shared entity could eventually make the interaction seem more coherent, but the idea depends on users accepting the privacy tradeoff Schiffmann described. For now, the pendant offers conversation, interruptions, and a set of personas the reviewer found difficult to tell apart.