Two numbers from a LogicMark national caregiver survey released yesterday are worth holding side by side.
Ninety percent of family caregivers show burnout symptoms. Twenty percent of them are in the severe category. And — the one that surprised us — roughly eight in ten say they would embrace AI-powered support.
If you've been in this category long enough, the first number is not news. The Cleo Family Health Index found 64% of sandwich- generation working women at high burnout risk. The AARP / NAC 2025 report put median caregiving hours at 27 a week. The trillion- dollar AARP figure is, in significant part, the invisible cost of that work. The 90% number lands harder, but it is in line with where every adjacent dataset has been pointing.
The second number is different. It says the audience is not gun-shy about AI in family care. The hesitation the category has been writing software around is, increasingly, our own hesitation — not the families'.
That changes the question.
The question is no longer "will families use AI"
For most of the last three years, AI in caregiving software has been framed around an assumption that families would be cautious. Soft-launch the AI features. Add a banner that warns about hallucinations. Let the user opt in cautiously, one feature at a time, after they've trusted the rest of the product for a while. The pace was slow because we assumed the audience needed time.
The LogicMark survey suggests the audience does not, in the aggregate, need that time. They want the help. The 8-in-10 number isn't qualified by "for low-risk tasks only" or "as long as a human is in the loop" — those caveats are ours, not theirs. Families looking at a calendar full of appointments, a medication list they didn't write, a stack of after-visit summaries no one has read, are saying: yes, please, help.
That changes the question from "will families use AI" to "what is the responsible shape of help they'll trust over time."
The trust burden, in three pieces
There are three things AI tools in caregiving have to get right to earn — and keep — the trust the LogicMark data says is already extended:
1. The patient stays in the loop. AI summaries that the family reads but the patient doesn't get to see are a category mistake. The Language Access for All Act moving through Congress puts this on a legal footing for LEP patients specifically, but the principle is universal: the person whose care is being coordinated is also entitled to read the coordination. Software that treats them as the subject of records, not a participant in their own care, breaks the trust the families are willing to give. We wrote about this last week under the heading who counts as "shared" in a shared caregiver workspace.
2. The source artifact is preserved. AI translation, AI summarization, AI lab-value extraction — every one of these produces a derivative artifact that the family acts on. The CHI 2026 LEP paper documented what happens when families act on AI output without the source: hallucinations get propagated, regional varieties of language get flattened, figurative content gets mistranslated as literal. The discipline is to keep the source — the doctor's original note, the English version of the message, the lab PDF — as the artifact of record, and treat the AI output as a bridge to it, not a replacement.
3. The off-switch is real. State AI-disclosure laws are moving fast. TRAIGA in Texas took effect January 1, 2026, requiring written, visible disclosure of AI use in clinical care. Colorado's AI Act enforcement begins June 30. California's AB 489 bars AI from implying a healthcare license. The pattern: the ability for the family to know AI is being used, and to turn it off, is moving from nice-to-have to legal floor. A product where the AI is on by default and the off-switch is buried five settings deep is, increasingly, a product with regulatory exposure.
The 8-in-10 number isn't an invitation to lean into the AI more loudly. It's an invitation to lean into the discipline more visibly.
What this looks like in the workspace
Kintaria's AI features ship off by default. Visit summaries, document translation, lab extraction, medication scanning — all disabled until the workspace owner turns them on, per feature, in Settings → AI Features. Until they do, no workspace content goes to any AI service.
When AI is on:
- Every AI-touched surface is labeled at the point of use — AI preview, machine-translated from English, AI-extracted from this document — not in the privacy policy.
- The English original (for bilingual surfaces) or the source document (for extraction) is preserved as the artifact of record. The AI output is shown alongside, not in place of, the source.
- The audit log captures which AI surface produced which row, with a back-pointer to the originating document.
- Per-vendor accountability is published on the security page — which third-party AI services touch workspace content, what BAA status each has, and what the workspace owner can do to opt out.
This is what we think the LogicMark data is asking for — specifically. Not "AI yes" but "AI on these terms."
The broader category
Family caregiving software is going to look very different in twenty-four months than it does today. The AI capability is real, the user appetite is real, and the regulatory floor is hardening fast. The companies that build for the families that already exist — including the patient as a participant, including the source as the artifact of record, including the off-switch as a first-class affordance — will be the ones the trust the LogicMark survey is offering actually flows toward.
The companies that read 8-in-10 openness as permission to ship the AI on by default, hide the disclosure, and bury the source will get a few quarters of growth and then a slow erosion as the state AI-disclosure laws compound and the first high-visibility hallucination story gets the category in front of a Congressional hearing.
We're trying to build for the first version. The data this week says the families are ready to meet us there.