Winning AI Device Owns Behavioral Data

The core bet in AI hardware is that the winning device will not just run a model, it will own the stream of behavioral data that teaches the model what matters to one specific person. Rabbit, Friend, Humane, and similar devices each chose a form factor that let them capture different slices of context, like voice, camera input, or ambient activity, because phones and app layer assistants usually cannot see enough across apps and moments to build that same user understanding.

• Rabbit and Humane pursued broad assistant devices, which gave them room to invent new interactions, but also forced them to solve hardware, distribution, and context gathering all at once. That made the product harder to get right than a phone based assistant that can piggyback on an existing device people already carry.
  1 sacra 2 sacra 3 sacra
• Friend took a narrower path. It centered on companionship, not general task execution, which reduced the workflow complexity, but also limited the data surface and the number of daily jobs the device could perform. By late 2025, it shifted away from dedicated hardware toward a web based chatbot, showing how hard standalone devices are to sustain.
  2 sacra 4 sacra
• The practical advantage of phone retrofits is distribution, but the practical advantage of new hardware is control. A custom device or OS can decide what signals to collect, when to listen, what to show on a tiny display or in audio, and how to personalize that output. That is why companies keep moving lower in the stack, even when it is much harder.
  1 sacra 3 sacra 4 sacra

The likely direction is that AI specific experiences survive, but many AI specific devices do not. The functions pioneered by pendants and pins are getting pulled into glasses, earbuds, watches, and AI aware phone software. The durable layer is the intelligence system that learns the user across contexts, then adapts to whatever hardware form factor wins distribution.