Worker ownership works.
AI makes it scale from day one.

Why co-ownership fixes home care

The home care sector loses ~64% of its workforce every year. Not because the work is undoable — because the structure is broken. Workers have no voice, no scheduling control, no stake in outcomes. They leave. The agency hires again. The family starts over. The cost compounds.

Worker-owned cooperatives are structurally different. When workers own the agency, they govern it. They set schedules. They mentor each other. They stay — not because they earn more, but because it is their organization.

"Workers at co-ops described feeling valued in ways that contrasted sharply with prior traditional agency experience — attributing this specifically to the cooperative structure, not to individual managers."

The model isn't new. It has been proven for fifty years.

In 1971, in San Francisco's Chinatown–North Beach, a physician and a social worker — Dr. William Gee and Marie-Louise Ansak — built something for aging Italian, Chinese, and Filipino families who wanted to grow old at home, together, rather than be institutionalized. They called it On Lok — Cantonese for "peaceful, happy abode." A small interdisciplinary team that talked every day and coordinated everything a person needed to stay home: a doctor, a nurse, meals, therapy, a ride.

On Lok became the national PACE program — and a half-century of evidence says it works: participants are far less likely to be hospitalized or placed in a nursing home, and report higher satisfaction than comparable populations. It asks the right question — not "who pays for this service?" but "what does this person need to stay independent?" Sometimes a physician. Sometimes a ride. Sometimes a refrigerator so the medication stays cold.

co-op.care keeps what On Lok proved — manage the person, not the coverage; the team that knows you; families kept whole at home — and sheds the weight that kept it rare. No day center. No capitation license. A cooperative, not a fifteen-million-dollar build. The same spirit, at cooperative weight, for the millions PACE will never reach.

These claims circulate widely in cooperative advocacy. Each was tested by three independent AI reviewers and failed 2-of-3 to confirm. We do not use them.

The constraint that kept co-ops small is gone

CHCA is the strongest proof point in the sector. It took 40 years and a South Bronx community development organization to build. For four decades, the operational reality of running a home care agency — scheduling, matching, compliance, billing, care coordination — required significant administrative infrastructure that small co-ops couldn't afford.

That constraint was real. It explains why CHCA is still one of very few large home care cooperatives in the country after 40 years of proving the model works.

AI eliminates that constraint. CareOS — co-op.care's operating system — handles scheduling, matching, care coordination, LMN generation, and overnight briefing. The same operational capability CHCA built over decades is available from the first day of operation. A cooperative of five caregivers can run with the administrative infrastructure of a scaled operator.

The economics of cooperative ownership were always better in home care. For the first time, a new cooperative can access those economics from day one — without a 40-year organizational build.

And the moat is exactly what the agentic age can't copy

The same wave of AI that removes the constraint also commoditizes the thing most startups mistake for their moat. In June 2026 Google shipped dynamic sub-agents as a default feature of its developer platform; the orchestration that looked like a product a year ago is now a checkbox. When the agent itself is free, the durable advantage is whatever an agent can't be. Four things qualify — and co-op.care is built on all four.

So the 2026 case has two halves. AI hands a five-person co-op the operations CHCA needed forty years to build — and, by commoditizing everything else it touches, it leaves the operator, the proof, the ownership, and the resonance as the only durable moats standing. We did not choose those four because they're fashionable. We chose them because they're the four an agent will never hold.

The honest answer: we have strong evidence the mechanism works, a 40-year proof point that it persists, and a specific thesis about why AI changes the scaling constraint. The Boulder pilot is the experiment that answers question one — and begins building the dataset for the others.