The rule, before the features
AI is good at the dull, bounded jobs — reading a label, laying out a standard structure, summarising a long note. It is not the thing that should decide what a person needs, what a clinical judgement is, or whether a risk is acceptable. Those belong to people who are accountable for them. So the rule we build to is one line: AI for admin, humans for care. Everything below is an application of it.
The reason to be loud about this is that the participant can't audit the software — they rely on the people around them, and those people rely on a tool that stays in its lane. A tool that quietly lets AI drift from admin into judgement is a tool that's offloading the part it was never qualified to do.
Where we use AI — three narrow, admin-only jobs
Reading a Webster pack to set up the record
Photograph a Webster pack and the AI extracts the medications to pre-fill a review — drug, strength, route, schedule. Then it stops. Nothing saves until a worker confirms each row; every row shows a confidence level and the raw label text the AI read, so a bad read is easy to catch; and the photo lives in the audit trail attached to the medication. The AI drafts the data entry. A human decides it's right. The photo is the evidence the human used.
Drafting the structure of a report
For an OT or behaviour support practitioner, Scrive drafts the scaffold of an FCA, AT or behaviour support report — the section headings, the standard format, the evidence prompts. The clinical reasoning, the recommendations and the plan content are written by the practitioner, in their voice. Scrive drafts the structure; the practitioner owns every clinical judgement and the plan content. It also tracks the interim and comprehensive due dates — but it does not lodge the plan to the NDIS Commission, it does not authorise restrictive practices, and it does not make anyone a registered practitioner. Drafting and deadline-tracking is the boundary; the lodgement and the authorisations stay with the practitioner.
Summarising a handover
A long shift note becomes a short, readable summary so the next worker gets the gist fast — with the full note still there underneath. The summary is a convenience layer over the record, never a replacement for it, and never a decision about the person.
The question isn't whether software uses AI. It's what it's allowed to decide — and in care, the answer is nothing.
Where we don't — and won't
The lines we hold
AI doesn't make care decisions. What support a person needs, how a risk is managed, what a recommendation says — those are human judgements, by people accountable for them.
AI doesn't write the clinical content. It can lay out a report's structure; the clinician writes every clinical sentence. Their registration is on the line, not a model's.
AI doesn't act on its own. Nothing the AI produces is saved, sent or shared until a person reviews it. The model proposes; the human disposes.
Why the boundary is the point
Holding AI to admin isn't caution for its own sake — it's what makes the time-saving safe to use. Because the human stays in the decision, the OT can let Scrive draft the structure and spend the saved hour on the clinical thinking that's actually theirs. Because the worker confirms every medication, the provider can let the camera do the typing without putting the Medication Administration Record at risk. The admin gets faster precisely because the judgement never moves.
That's also why we name where the AI is and what it saw — the confidence level, the source label, the photo in the trail. Honest AI in care shows its working so a person can check it. Software that hides the seam between what the AI did and what the human decided is the software to be wary of.
AI for admin. Humans for care.
The dull jobs get faster; every decision stays with a person. That's how Clearline uses AI across Aura OS, Pilot, Compass and Scrive — Australian-hosted, on Clearline Connect.
Questions
Does Clearline use AI to make care or clinical decisions?
No. AI is used narrowly for admin — reading a Webster pack, drafting a report's structure, summarising a handover. Every care decision and clinical judgement stays with the human, who reviews and owns the output.
Where does Clearline use AI?
Three admin-only places: Webster pack OCR (the worker confirms every medication), structural drafting of FCA, AT and behaviour support reports in Scrive (the practitioner writes every clinical sentence), and handover summaries in Aura OS. Each shows its source so a human can check it.
Does Scrive's AI write the behaviour support plan?
No. Scrive drafts the structure; the practitioner owns every clinical judgement and the plan content. It tracks due dates but does not lodge the plan, authorise restrictive practices, or confer registration.
Is AI a risk to medication accuracy?
AI pre-fills the record from the Webster pack photo, but nothing saves until a human confirms each row — with a confidence level, the raw label text, and the photo kept in the audit trail.