The human part2 min read
The parts we won't automate
A short list of things a machine should never be first to do, and why the software is better for having the list at all.
Nestaid ·

There is a version of this technology that ends badly, and it isn't hard to picture. A family calls at a frightening moment and reaches something that will never hand them to a person. A caregiver is texted at one in the morning about a shift forty minutes away, by a system that doesn't know she has a child asleep down the hall. A decision gets made somewhere in the building and nobody can say why.
None of that requires bad intentions. It only requires building the automation first and the limits afterwards.
Legwork is not judgment
What our agents do is legwork. The Coordinator ranks caregivers by travel, declared availability, authorization and history, and offers a shift in small waves. What it does not do is decide that this client, in this particular month, needs the caregiver who has known her for two years. That is a judgment about a person, and it belongs to a person who knows her.
The same line runs through the rest of the roster. The EVV Expert calls about a missed clock-out and writes the reason onto the visit; what to do about a caregiver who is missing them every week is a manager's conversation. The Care Plan Generator drafts from the service code; a clinician reads it, changes it and signs it.
A ladder, not a switch
Every agent can start at the bottom of an autonomy ladder: it proposes and waits. It drafts the list and the message, and nothing sends until somebody in your office says so. When the office is ready — once they've read enough of its work to have an opinion — it moves up a rung, one agent at a time. It can be moved back down the same afternoon.
Trust isn't a setting we can ship with. The software has to earn it from your office, one week at a time.
The limits that are built in
Some things aren't a rung on the ladder at all. They're fixed, because a promise an office can switch off on a bad Friday isn't a promise:
- No client's name in a coverage message. The shift and the distance, nothing else.
- Declared availability honored on every run, and declining is one tap that is remembered.
- Location captured at clock-in and clock-out for the visit, not in between.
- No double-booking — for people or for agents.
- One audit log, covering both.
What the hours are actually for
Taking the phone calls away is not a way of getting people out of home care. Nobody got into this work to be a switchboard. Give a scheduler back the forty-five minutes it takes to fill one Saturday shift, and give an owner back the nights, and the hours go where they were always supposed to go: into the visit, the family, and the caregiver who is thinking about leaving and could be talked out of it by somebody who has the time to notice.
The other two
The shift
Recording the work is not doing the work
Home care software has spent fifteen years getting better at remembering. The interesting question now is what it actually did today.
A night, from four seats
One call-out, four people, one record
A Saturday 7 a.m. loses its caregiver at 2:47 in the morning. Here is what each person sees, and why they are all looking at the same thing.
