A design diary · Azure · Aged care · Agentic AI
Agents That Know When to Stop
Designing agentic AI for aged care, where a person always makes the call.
Imagine an aged-care operator with 180 facilities and home-care stations and about 9,000 caregivers, who spend up to a third of every shift on paperwork. This blog follows the design of five AI agents that hand that time back. Each one stops at the point where a person should decide.
Start with Part 1 →Four rules that shape every design decision
No automated care decisions
Agents draft, recommend and alert. Nurses and care managers decide.
Health records stay in-country
Identifiable data never leaves Australia.
Only de-identified metrics go regional
A regional view of quality, with no names attached.
Night alerts work offline
Falls alerting runs on site and doesn't wait for the cloud.
Latest posts
Three boundaries: facility, country, region
Where data is allowed to live, what may cross each line, and why a storm at 2 a.m. shouldn't matter.
Part 1 · Post 2The whole system on one page
Seven principles, one diagram, and the sentence that holds it together: the model drafts, rules decide, people confirm.
Part 1 · Post 1What problem are we designing for?
180 sites, 9,000 caregivers, up to a third of every shift on paperwork, and four rules that can't be broken.
The series
Four parts, from the big picture down to each agent and back out again.
Meet the night shift
Each post opens with a scene at Zava Care Group. These people are fictional.
Registered nurse, night shift
Makes the clinical calls. Every agent ends up at her desk or her handset.
Personal care worker
Does the hands-on care, and the paperwork that follows it.
Resident, 88
Likes the garden, dislikes fuss. She owns the record that the agents work from.
Eleanor's daughter, registered supporter
Lives two hours away and wants to know how Mum's night was.
Facility manager
Approves the roster and is answerable for the facility's compliance.