OpenAI, Jony Ive struggle with technical details on secretive new AI gadget (arstechnica.com)

OpenAI and designer Jony Ive (whose firm io was bought by OpenAI for $6.5B) are racing to ship a secretive palm-sized, screenless AI gadget that uses cameras, microphones and speakers to take audio‑visual cues from the environment and respond to users. Despite completed industrial design work, teams are encountering unresolved technical and infrastructure problems that could delay launch: software integration, how to define the assistant’s “personality,” privacy and continuous‑sensing policies, and — crucially — where and how to run the heavy models that power its intelligence. The challenges matter because the device is reported to be “always on,” continuously building an assistant “memory” from sensor data, which raises serious privacy, storage and inference-cost questions. OpenAI must balance model size, latency and costs—deciding between local edge compute, costly on‑device acceleration, or cloud inference at scale—while competing with Amazon and Google on backend capacity. If solved, the product could advance ambient, multimodal consumer AI; if not, it highlights core industry friction points: compute provisioning, real‑time multimodal perception, personalized memory systems, and managing privacy/UX tradeoffs for always‑listening devices.