The first wave of clinical AI was about documentation. Ambient scribes that listen to a consultation and write the note have spread quickly, and for good reason — they hand clinicians back time that paperwork had quietly taken. But documentation is becoming table stakes, and the contested ground is moving to something harder and higher-stakes: the clinical decision itself.
What an AI scribe does — and why it is commoditising
An AI scribe transcribes a consultation, structures it into a note, and often drafts letters and summaries. The value is immediate: less time typing, more time with the patient, fewer evenings finishing records. Adoption reflects that — Heidi alone reports its scribe has supported over 100 million clinical interactions globally, and Tortus is deployed across thousands of UK practices.
But the underlying capability — turning speech into structured clinical text — is now offered by many vendors with converging features. When several tools do roughly the same job to a similar standard, documentation becomes a baseline expectation rather than a differentiator.
The tell: everyone is climbing from scribe to decision support
The market's direction is visible in the product roadmaps. The leading scribe companies are not staying scribes:
- Tandem moved from AI scribe to coding assistant to clinical decision support, each step CE marked as a Class IIa device.
- Heidi moved from scribe to "AI Care Partner", launching Heidi Evidence, a decision-support and evidence layer.
- Tortus is widening from transcription into coding suggestions, with a higher regulatory class reported as pending.
For the companies building in this space, the scribe is increasingly an entry point into wider clinical workflow support — not the finished product.
Why decisions are a different, harder problem
Documentation and decision support sit at different points on the risk curve.
| AI scribe | Clinical decision support | |
|---|---|---|
| Core job | Documents the encounter | Informs a patient-specific decision |
| Typical failure | Mistranscription, caught at sign-off | Subtle bias in reasoning, harder to catch |
| Risk profile | Lower | Higher |
| Regulatory tendency | Often Class I | Generally Class IIa (EU MDR Rule 11) |
A scribe that mistranscribes a sentence is usually caught when the clinician signs off the note. Decision support is subtler: it offers information that shapes how a clinician reasons, and a quiet bias or omission is harder to spot and carries more weight. That is exactly why decision support attracts a higher regulatory classification — and why it deserves more scrutiny, not less.
The frontier, then, is not "can the tool write the note?" but "can it reduce uncertainty around the decision — safely, transparently, and without quietly taking the decision away from the clinician?"
iatroX's entry point: the knowledge and reasoning layer
iatroX did not set out to build a better scribe. It works the layer underneath the decision: the knowledge a clinician draws on, and the reasoning they apply. AskIatroX reduces uncertainty by surfacing fast, source-linked answers from validated UK guidance — NICE, CKS, SIGN and the SmPC — so a clinician can check a threshold, a pathway or a dose against a cited source in seconds.
There is a second half that documentation tools do not touch: learning. iatroX's adaptive question banks and Socratic tutor are built to train clinical reasoning over time, across the exams and specialties clinicians actually sit. A scribe captures the decision you made today; the knowledge-and-learning layer improves the decisions you make next year.
AskIatroX is a UKCA-registered, MHRA-listed Class I device — it informs the clinician's judgement rather than automating or directing it, which is why it sits at Class I rather than the Class IIa of in-consultation decision engines. That is a different job, not a smaller one.
Different centres of gravity, not a head-to-head
It is more accurate, and more useful, to see the centres of gravity than to frame every tool as a rival. Scribes are anchored to admin reduction. In-consultation decision support, as Tandem is building it, is anchored to the documented encounter and the EHR. iatroX is anchored to trusted answers and clinical learning, available before, during and after care, independent of any record system, with the core clinical reference free.
None of these cancels the others out. A clinician might dictate a note with one tool, check a sourced answer with another, and revise for an exam with a third. The mistake is to assume they are interchangeable.
So when a new clinical AI crosses your desk, the sharper question is not "is it good?" but "what is it for?" Documentation was the first battle. Decisions — and the knowledge behind them — are the next one.
Frequently asked questions
Are AI scribes and clinical decision support the same thing? No. A scribe documents the consultation; clinical decision support offers information that shapes a clinical decision. They sit at different points on the risk curve, which is why decision support generally attracts a higher regulatory classification.
Why does decision support carry a higher regulatory class than a scribe? Because it can influence the decision itself, not just the record of it. Under EU MDR Rule 11, software providing information used for diagnostic or therapeutic decisions is generally Class IIa, whereas pure documentation carries a lower risk profile.
Are AI scribe companies moving into decision support? Yes. Tandem, Heidi and Tortus have all extended beyond transcription — into coding and, in Tandem's and Heidi's case, decision-support or evidence products — indicating that scribes are becoming an entry point into wider workflow support.
Does iatroX offer an AI scribe? No. iatroX works the knowledge, reasoning and learning layer rather than the documentation layer. It is complementary to scribes — a clinician can document with a scribe and use AskIatroX for fast, cited clinical answers and exam learning.