AI in Medicine — with Panacea AI.

Why this episode now.

Clinical artificial intelligence has moved from "interesting in theory" to "in the consulting room next door" faster than most surgeons predicted. By May 2026, the AI medical scribe — an ambient-listening tool that drafts the consultation note in real time — is in routine use in a meaningful slice of Australian general practice, and is starting to appear in specialist rooms. Tools that pre-read shoulder MRIs, that flag radiographs for fracture, that re-rank an incoming worklist by clinical urgency, are no longer hypothetical. The question is no longer "will AI come into clinical work" — it is "what does the practice actually do with it, and what does the patient need to know."

Episode 2 is the conversation. The format is the two of us in the rooms, with a guest from Panacea AI — one of the Australian clinical-AI companies whose work sits inside the scribe / clinical-documentation space. The point of the episode is not to sell anything; it is to map honestly where this technology is, what it actually does, and what it does not do.

What clinical AI is in May 2026 — the lay of the land.

Useful to separate four distinct layers of clinical AI in the wild right now, because they have very different evidence bases, very different risks, and very different consent implications:

1. The AI medical scribe. An ambient tool that listens to the consultation (with consent), drafts the clinical note, structures the history-examination-impression-plan format, and generates the referral letter or post-op letter as a by-product. The clinician edits and signs. The signal here is reduced after-hours documentation time — published Australian general-practice data have shown evening note-writing burden dropping substantially.
2. AI-assisted image interpretation. Models that pre-read radiographs, ultrasound, MRI and CT, flagging fractures, masses, and time-critical findings before the radiologist reaches them. Currently used as a triage layer, not a replacement for the radiology report. TGA-approved tools include Annalise.ai (Australian-developed, chest and head imaging) and several international platforms.
3. Clinical decision support. Large-language-model-based tools that summarise a complex chart, surface the relevant guideline, or rephrase a clinical question. These sit closer to "smart search" than "diagnosis"; the published evaluations are mixed and rapidly evolving.
4. Workflow and operations AI. Booking optimisation, no-show prediction, theatre-list re-ordering, claim-coding suggestions. Unglamorous, increasingly common, and where the productivity story is currently strongest.

Most of the conversation in this episode is about layer 1 (the scribe) and the patient-facing implications of all four.

What Panacea AI is, and why we had them on.

Panacea AI is an Australian clinical-AI company working in the medical-scribe / clinical-documentation space. We had them on because the conversation about clinical AI in 2026 is dominated by overseas vendors whose product was not built around Australian clinical realities — MBS item descriptors, the specific shape of a Queensland GP's day, the WorkCover and CTP paperwork, the Medicare referral conventions. A local builder with skin in the game has a different perspective on what is actually buildable for an Australian practice, and we wanted that voice on the record.

This is not an endorsement and this is not a sponsorship — the episode is a conversation, not an advertisement.

What the conversation keeps coming back to.

• Consent and the patient experience. The medical scribe is recording the consultation. The patient has to know, consent has to be informed, and the recording has to be handled in a way that aligns with the Privacy Act and the AHPRA Code of Conduct. Done well, patients overwhelmingly prefer a consultation where the clinician is looking at them rather than at the screen.
• The note still belongs to the clinician. The AI drafts; the clinician edits, signs, and is responsible. The medico-legal liability does not transfer to the model. Anyone who treats the AI-drafted note as a finished product is misusing the tool.
• Hallucination is real and shrinking. Earlier-generation scribes occasionally inserted findings that were not in the audio. Current-generation models with Australian clinical training data hallucinate less but the risk is non-zero. Editing matters.
• The AHPRA position. AHPRA released a position statement on AI in clinical practice (the 2024 version, updated since) that places the responsibility on the registered practitioner. The clinician using AI is responsible for the output; the tool is an aid, not a substitute. The published guidance is straightforward and worth reading.
• What this changes for shoulder surgery specifically. The scribe lets the surgeon spend more of the consultation actually looking at the patient and at the imaging, less of it typing. The radiology AI pre-read can flag a high-grade cuff tear or a Hill-Sachs lesion before the formal report lands. Neither replaces the surgical judgement — both buy more of the surgeon's attention for it.

The skeptic's case — and the optimist's case.

The skeptic's case is reasonable: the technology is moving faster than the regulatory framework, the consent process is harder than the marketing suggests, hallucination has not been completely solved, and the economic incentives of AI vendors do not always align with the clinical incentives of the practice. The Australian primary-care sector is in the middle of a slow-motion absorption of all of this and the early adopters carry risk the late adopters do not.

The optimist's case is equally reasonable: the documentation burden in Australian medicine is a measurable cause of clinician burnout, the early Australian data on scribe adoption show real time savings, the radiology pre-read tools demonstrably catch findings that humans miss in fatigue states, and the practice that integrates these tools well in 2026 has a four-year head start on the practice that does not.

The honest position sits between the two. The episode does not try to land on either side. It tries to make the trade-offs visible so the clinician and the patient can have the conversation deliberately.

For patients.

If you have a consultation with Dr Coory and an AI scribe is being used, you will be asked first. The recording is not retained beyond the period needed to draft the note. The note is reviewed and signed by the surgeon. If you prefer the consultation without the scribe, that is fine — the consultation will run the same way without it.

Episode list

• Ep 1 · Welcome to UnscrubbedThe format, the combined-clinic logic, and the questions patients actually ask →
• Ep 2 · AI in Medicine with Panacea AIThe AI scribe, radiology, decision support, and the AHPRA framework. You are here.
• Ep 3 · Healthcare Hot TopicsRound-up format: GP workforce, Medicare, AI in practice, hospital pay deals →

---

Related on this site

• The Clinical Experience — the combined-clinic model
• Before I look at your imaging — the consultation as the first treatment
• The full video library
• AHPRA — the official regulator