I have been a caregiver for fifteen years. I have cared for my mother through dementia. I have supported three neighbours through cancer treatment. I have coordinated care for children with complex needs in a school that was chronically underfunded and understaffed.
Nobody in that time ever asked me what I needed. They asked what the care recipient needed. That distinction matters.
What AI-assisted care gets wrong
The proposals I keep seeing for AI in caregiving follow a pattern. They start with a real problem — scheduling, information gaps, isolation — and then propose a technical solution that addresses the symptom while missing the cause entirely.
An AI scheduling system doesn't solve the problem that there aren't enough care workers. It makes the shortage more efficient. An AI that reminds an elderly person to take their medication doesn't replace the relationship that makes them feel like taking it is worth it. A chatbot that checks in on someone living alone doesn't provide what a phone call from a neighbour provides.
This isn't technophobia. I use technology. I am writing this on a laptop, communicating it through a platform. The point isn't that technology can't help with care. It's that care is fundamentally relational, and relationships cannot be automated.
The data problem
There is also a deeper issue. To train AI systems on care work, you need data about care work. That data is personal — it involves the most intimate moments of people's lives. When a company proposes to "digitise" care coordination, what they are actually proposing is to turn the relationship between a care worker and a care recipient into structured data that can be used to train models that will be sold to other care organisations.
The care recipient didn't consent to that. The care worker didn't consent to that. But the contract is buried in the software license of the scheduling tool, and nobody read it.
What actually works
What actually works in care is time, continuity, and recognition. It works when care workers are paid properly and not burning out every eighteen months. It works when communities have genuine infrastructure — drop-in centres, respite services, accessible transportation — rather than an app that simulates those things.
Technology that helps with care should be minimal and invisible. It should reduce administrative burden so that the time freed up goes back into relationships. It should never collect data beyond what is strictly necessary for the immediate task.
The Neighbourhood Mesh Care idea in the Idea Bank gets this right. No central database. No AI. Just people, coordinated.