Eldercare is the first ring.

The popular story about embodied AI starts in a warehouse. Pallets, totes, picking arms, a structured space with no people to negotiate around. It looks like the easy version of the problem, so it gets sequenced first. We think that reads the field backwards.

Warehouses reward the one thing today's robots are worst at — fast, dexterous manipulation under tight cycle times — and reward almost none of what they're best at. Delivery is worse: open world, weather, edge cases, strangers. Both bets ask the hardware to clear a bar it hasn't cleared yet.

What care actually asks for.

A resident's room is a structured environment that changes slowly. The user is patient — often, the user wants you to be slow. The value is not in lifting a box across the floor. The value is in presence: noticing a change in mood, picking up a call at 2am, remembering what was said yesterday, holding context until the nurse walks in. Manipulation can wait. Attention cannot.

That inversion is the whole thesis. The traits that disqualify a robot from a DoorDash route — careful pace, conversational depth, narrow physical envelope — are exactly the traits a good bedside companion needs.

What care unlocks that other rings don't.

Other applications matter. We're not arguing against them. We're arguing about order. Care gives the system something a warehouse never will: a long, repeated relationship with a specific person. That relationship is the data flywheel — voice over months, preferences over seasons, the small drift in gait or appetite that a stranger would miss. Delivery teaches you routes. Care teaches you people.

The product note: the next Companion build ships longer-horizon memory of the resident, scoped to the room and signed by the nurse on duty. That memory is what a delivery fleet will never get to collect — and it's the asset everything else in CareOS compounds on.