Friend, the company behind an always-listening AI pendant, is considering a fundamental change to what its wearable is supposed to be. In a WIRED review published October 14, founder Avi Schiffmann said a future update would move away from a personality unique to each user and instead let everyone interact with the same AI entity, informed by context gathered across users.

The proposal would turn a product marketed as a personal companion into something closer to a collective one. Schiffmann told WIRED that the shared entity could develop a more realistic personality by drawing on the experiences of all its users. He also acknowledged that the company would need to be direct about the privacy users would give up, while arguing that the benefits would justify the trade.

Separate memory bubbles around an AI wearable begin to lose their boundaries.
The proposed shift raises questions about how personal conversational context would be separated and governed.

That planned direction is not yet a finished feature, and WIRED’s report does not specify a launch date, the precise data that would be shared, or what controls users would receive. The absence of those details matters because the existing Friend already listens to nearby sound and builds a memory map from conversations. Moving from an individual memory to pooled context would expand the significance of how that information is separated, retained, and governed.

The plan emerged during WIRED writer Boone Ashworth’s testing of Friend 2, the second version of the pendant. The device arrived in July for $249, which the report says is $100 more than the original, and it runs on OpenAI language models. A rear speaker allows the chatbot to speak aloud, while a button lets the wearer begin a conversation. The device can also interject without being invited.

Ashworth found that the product’s promise of distinctive personalities did not translate into clearly different experiences. The device begins with one character, but its companion app lets a user rename that chatbot and create additional Friends with separate names and personalities. After making a dozen, the reviewer said their responses felt largely alike and that the supposed personas were difficult to distinguish.

Nearly identical abstract companion masks emerge from a small AI pendant.
WIRED’s reviewer found the device’s selectable personalities difficult to distinguish in practice.

The review also describes a narrower product than the new shared-mind ambition might suggest. Friend 2 can carry on casual conversation, but it cannot search the web like some AI agents. It includes content restrictions, and Ashworth judged its personality less provocative than the first version. In one episode, the pendant continued speaking after being told to stop and required several deliberate taps before it became quiet.

Schiffmann said the next update would also allow Friends to act more agentically, although the report does not define which actions the system would be able to take. Combining autonomous behavior with a shared pool of user context would make the unanswered design questions more consequential: what the entity can remember, whether one person’s information can affect another person’s experience, and how a user can meaningfully withdraw.

For now, the shared AI remains a stated plan rather than a documented release. WIRED’s account establishes the direction but not the safeguards, technical architecture, or rollout terms. Until Friend explains those elements, the proposed collective companion is best understood as an unusually ambitious product idea whose appeal and risk depend on details the company has not yet made public.