Whether Heidi makes it or not as a global AI clinical workflow layer provider, the story they are now telling about technology will.
Tomorrow, your morning starts before your first patient arrives.
Your screen shows a dashboard: today’s list, open tasks from yesterday, and a pre-charted summary for each patient drawn from their longitudinal record. Not notes you wrote last time. A prepared brief – problems, medications, relevant history – assembled while you were driving in. You haven’t touched the keyboard yet.
That is just a little of what Heidi II is offering. Not as a concept. As a demonstrated product, shown in a live launch event on 30 September 2026 and confirmed in hands-on demos afterward.
In a simple phrase put by founder Dr Thomas Kelly last night, and sure to resonate over the next year or so: “give the computing back to the computers”.
It was all demos of course, but if you work with AI you know it’s all doable now. It’s going to train and learn and iron out the wrinkles quickly.
Being a doctor is about to change, for the better, radically… Heidi II or not, but Heidi II is telling the story first and telling it pretty well.
Before the first patient: the pre-charted day
The centrepiece of Heidi II’s new home dashboard is what they call the pre-charted day.
Before you arrive, Heidi has pulled each patient’s record, summarised the relevant clinical history, and flagged open tasks.
In the demo, this was shown working across practice management systems – not just systems Heidi has deep integration with. For GPs on Best Practice in particular, Heidi uses robotic process automation (RPA) to do the pulling: it uses an AI agent to operate your screen the way you would, extracting what it needs from the record and preparing the brief.
RPA is not as smooth as native integration. Heidi admitted this. There are things a fully integrated system – like Lyrebird running inside Best Practice – will still do that RPA cannot.
But the pre-charted brief does not require deep integration to be genuinely useful. The question is whether a prepared patient summary waiting for you before each consult is worth running two systems in parallel.
If they can get it smooth, for a lot of GPs, it feels like a trade-off that might be worth making.
During the consult: agents working while you work
A more significant change is what happens while you are with your patient.
Heidi’s new agentic layer means that while you are in the room, Heidi can be outside it – drafting the referral letter for the patient you just saw, filling a PDF form complete with appropriate letterheads, scanning for a reply to an email that came in, pulling a full-text journal article to check the dosing you were uncertain about.
Heidi’s new medical knowledge layer integrates full-text access to the New England Journal of Medicine, JAMA, Wiley, and Cochrane. When Heidi retrieves evidence, it cites the specific paper, with a traceable link.
In the near term, Heidi is shipping what may be its most practically significant feature: a dedicated parallel browser that runs Heidi’s agents on a separate virtual computer alongside your own.
At the demo, the team showed 13 draft emails completed in just over a minute while the clinician was handling a separate queue on their own screen.
At the moment, RPA requires your own screen – meaning you can’t use your computer while Heidi is running its routines. The parallel browser removes that constraint. When it ships, the clunkiness that currently limits RPA’s appeal will likely start to go away.
After the consult: referrals, messaging, memory
One announcement from the post-event demo is worth singling out: Australia’s dominant secure clinical messaging network is now accessible from within Heidi. A GP can dictate a referral letter and send it as a secure message to any specialist or service in the HealthLink directory at a voice prompt.
Someone in the demo room asked the question directly: “So I could send my letter straight from Heidi rather than separately using HealthLink?”. The answer was yes.
For a GP who currently moves between their clinical software, HealthLink, and whatever they use to dictate or draft letters, the workflow implications are significant.
Related
The referral loop – consult, dictate, draft, review, send – collapses into a single agent instruction.
You see the next patient. Heidi handles the paperwork… the computer gets the computer work.
Heidi II also retains preferences you state in conversation – preferred summary formats, documentation style, shortcuts you use repeatedly – without you ever touching a settings page. The system learns your working style as a byproduct of using it. In six months of regular use – AI training on you – Heidi will become quite personalised to your style.
The Caveats
None of this will be free of trade-offs.
If you are on Best Practice and you want the deepest possible integration with your clinical record – the kind where every Heidi action is natively written back into your EMR without a workaround and in the knowledge of total compliance and safety – Lyrebird still has that advantage. Best Practice has invested in Lyrebird specifically to offer that integration.
Heidi’s RPA approach is a solution to the lockout problem, but it is not the same thing as native integration and Heidi acknowledges this. The question is whether the gap is wide enough to choose Lyrebird over Heidi’s far larger feature current set.
On pricing: the agentic features are free for the first month for clinician-tier subscribers. After that, a credit-based add-on applies, but Heidi has not yet set the price.
Their stated position is that they will absorb compute costs while they work out the economics. Agentic AI – agents running in parallel, drafting multiple documents, pulling full-text journals – uses substantially more compute than a scribe. What the ongoing cost looks like is genuinely unknown, and it will matter.
There is also a data question that came up at the demo. Gmail and Outlook integration exists – Heidi can scan your inbox as part of an agent workflow – but email data goes to Google or Microsoft servers as part of that process.
Heidi is using an model context protocol-based approach, and the data privacy implications depend on what those connectors do with the content. For practices with strict data governance requirements, this is worth clarifying before enabling. But the rule is pretty simple: don’t dictate something onto email that you would never put on email anyway.
Is AI for you?
If you are a GP who has been watching Heidi from the sidelines – interested in AI scribing but not sure whether the integration problem made it viable for a Best Practice practice – last night materially should have changed the answer.
Whether it’s Heidi that takes you to the future workflow described above or another software vendor, that future is coming, and you can’t see a world where any doctor doesn’t use it: give the computer work back to the doctor and be a real doctor again.
It’s literally back to the future.
The integration problem is not fully solved. But it is navigable in a way it was not six months ago, and the feature set you get in return is now considerably larger than anything a local competitor is publicly offering.
If you are already using Lyrebird and happy with the Best Practice integration, the calculus is harder. You would be trading a smoother record integration for a broader agentic capability set, at a pricing point that remains unknown.
And if you are a hospital doctor working inside Epic or Oracle: the case for using Heidi alongside your enterprise system, rather than waiting for that system to catch up, just got meaningfully stronger. Heidi’s speed of development and agility in innovation is something the incumbents genuinely cannot match.
The industry implications of all this – what it means for the software vendors, the scribes, the secure messaging companies – are addressed in the companion piece HERE.
