Home care agencies split on letting AI act on its own

Home care companies are testing AI that can take action on its own, and they do not agree on how fast to go. One agency is planning AI helpers for every job. A health care lawyer warns that a mistake in this field could mean severe harm or even death. Most voices in the story land in the middle: start with paperwork, not patient safety.

What this kind of AI is

The tools in question are called agentic AI. According to Stanford University, these systems can make decisions, plan and carry out actions, use other tools, and act on their own or partly on their own. That is different from the chatbots most people know. Those tools, called generative AI, write text or make images when asked, but they cannot finish tasks independently.

Who is moving ahead

TheKey, a home care company based in Delray, Florida, that works in 29 states and Canada, wants each employee role to have two to three AI agents in support. Its chief people officer, Emily Kraeter, said this could free caregiver managers to spend more time training caregivers and building relationships. TheKey is rolling out Caribou's Conductor scheduling product, which it expects to help find coverage when a caregiver calls out.

That scheduling effort began as a small project. Chief operating officer Greg Meyers said it became the start of a company-wide AI plan. Modivcare, a Denver-based company in 35 states, also sees a use: AI agents can help streamline call queues, according to Marshalina Ramos, its CEO of personal care services.

Why some are cautious

Greg Leighton, a privacy and incident response leader at the law firm Polsinelli, said agentic AI puts operational risk in the hands of the tool. The biggest risk is that the agent gets something wrong. He gave a hypothetical: a patient-monitoring tool that missed a serious medical issue or misjudged how urgent it was. In home health, he said, things can get high stakes fast.

Ramos pointed to the industry's heavy regulation and the federal crackdown on fraud, waste and abuse. She said providers want proven technology and that Modivcare has no appetite to be first to market. Modivcare is going after easier wins first to improve efficiency.

Where experts see a safer start

Leighton suggests agencies first use generative AI to improve existing processes and work through compliance questions. Agentic workflows can come afterward. He named several sensible early uses:

• Streamlining operations
• Coordinating appointments
• Billing
• Reimbursement management
• Other back-office administrative tasks

What to ask

For families and business owners alike, the useful question is where an AI tool decides something alone and what happens if it is wrong. Ramos put the test simply: new technology should not disrupt patient care. The story does not report results from any of these rollouts yet, so how well the tools work in practice remains unknown.