Use clinical automation deliberately: retain an account of the patient's problem, check assumptions that could change care and practise challenging or recovering unfinished work. The aim is not to repeat every automated step. It is to remain able to recognise when a fluent proposal, an apparently completed task or an inherited interpretation no longer fits.
This is a professional-use framework, not a Heidi II configuration manual. Heidi's launch materials checked on 29 September 2026 say the new capabilities are not available in the UK or EU. Eligible users elsewhere should follow the permissions and instructions applicable to their own deployment; UK readers should not treat the international launch as local availability.
Retain a short independent account of the problem
Before accepting a prepared interpretation in an appropriate case, identify the main uncertainty and what would change the plan. This need not be a second full consultation or a written essay. Its purpose is to preserve an understanding against which the proposed work can be assessed.
For a simple administrative action, the relevant question may be whether the patient has already completed the agreed step. For a more complex encounter, it may concern a change in circumstances, conflicting information or an unresolved clinical interpretation.
This approach should be proportionate. Requiring elaborate independent reasoning before every routine transmission would be burdensome without necessarily improving judgement. The focus belongs where the action depends on an assumption that could reasonably be wrong.
Check meaning before polishing wording
Consider a fictional follow-up request prepared after a consultation. Its wording is professional, but it assumes that a previous referral was received and accepted. The record only confirms that a draft was created.
A grammar-focused review would miss the problem. A meaning-focused review would distinguish preparation, submission, acknowledgement and acceptance. Before approving further work, the clinician would establish which state actually applies and what remains uncertain.
The same principle applies to identity, chronology and source attribution. A statement supplied by the patient, a finding observed by a clinician and an interpretation generated from the record should not silently acquire the same status. Review should concentrate on changes that affect the task, not on making already clear sentences more elegant.
Match supervision to the consequence
Heidi's 29 September 2026 product introduction describes Manual mode, a selectable Auto mode and limits on permitted actions. These are published design features, not a guarantee that every workflow suits the same level of supervision.
A practical deployment should identify where authority changes. Preparing material is different from authorising it to leave the system. Updating a draft is different from changing the clinical record. Scheduling a repeated process is different from approving an individual case after reviewing its context.
The organisation should be able to explain who can change those permissions and how staff will recognise when a task has moved beyond the agreed scope. The aim is not maximum friction. It is well-placed friction where a mistaken assumption would have a meaningful consequence.
Practise disagreement before it becomes urgent
An approved simulation can contain a plausible proposal with one important omission. The clinician's task is to identify the missing assumption, explain why it matters and change the proposed work. Another case can present a correct proposal, testing whether the user avoids unnecessary rejection.
A recovery exercise should begin after partial completion. The participant should establish what happened, what remains uncertain and what must not be repeated. Including a colleague who did not initiate the task tests whether the handback is understandable beyond the original user's memory.
These are proposed training exercises, not claims about an existing Heidi-specific curriculum or results we have observed. Deliberately flawed material belongs in a suitable educational environment, not in live records or patient communications.
Protect learning outside the live workflow
Independent attempts, expert examples, feedback and later application can be combined without pretending one method is universally superior. The amount of guidance should fit the learner's experience and the difficulty of the task.
A small randomised medical-education study published on 19 January 2026 illustrates that nuance. In CREST, novice students receiving example-based teaching performed better than those receiving self-explanation-based teaching at the earlier assessment, but the difference was not significant at the later assessment. This was ophthalmology education, not a trial of AI agents. The original study supports attention to instructional design rather than the slogan that more unaided effort always produces better learning.
For a clinician reviewing an assisted output, a useful educational follow-up might involve explaining why an alternative was rejected, then applying the same distinction to a different case later. The learning objective should remain visible instead of equating time spent reading an answer with competence gained.
Where iatroX fits, and where it does not
This article is published by iatroX and includes iatroX among the options discussed. A clinical workflow agent and a separate learning platform serve different purposes; neither should be presented as a universal replacement for the other.
Per iatroX's September 2026 product description, question banks offer adaptive sequencing and spaced repetition, while the Socratic Tutor opens on the attempted question and uses follow-ups to explore misconceptions. iatroX Simulations provides clinician-reviewed cases with voice and text, practice and exam modes, and transcript-linked feedback. One complete simulation is available free. These are educational features, not evidence that a subscription prevents deskilling, predicts examination success or certifies competence to supervise Heidi II.
Ask-iatroX and free question access are genuinely free, with no trial expiry or verification gate. The paid September 2026 subscription combines question banks, Socratic Tutor, study planner, Simulations and CPD tools for £99 paid upfront annually, equivalent to £8.25 a month billed annually, or £29 monthly. Simulations and CPD are included rather than separate add-ons. Three monthly payments total £87 and four total £116, so annual billing costs less from the fourth month.
The value is several learning methods supporting one relevant professional goal. It is not access to unrelated examinations, nor a claim that paying for education automatically makes a clinical agent safe.
Choose the next step by the problem you need to solve
For an eligible clinician seeking to reduce a defined administrative burden, evaluate the actual Heidi II workflow and its local permissions. For someone checking a general clinical question, a source-linked reference session may be more appropriate. For a learner who needs to understand why an answer is wrong, question-based discussion, supervised teaching or simulation addresses a different need.
CPD evidence also needs the right interpretation. Per iatroX's September 2026 product description, learners review and personalise draft records, with PDF export and direct FourteenFish export for linked accounts; completed evidence remains accessible after the subscription ends. A professional learning record is not automatically formally accredited CME or proof of independent competence.
The organising principle is simple: remove work that does not need your attention, but preserve the understanding that tells you when your attention is needed.
Frequently asked questions
Can I use Heidi II in the UK now?
The launch materials checked on 29 September 2026 exclude the new capabilities from the UK and EU. Existing Heidi products have their own availability and should be considered separately.
Does preserving judgement require checking every automated click?
No, the objective is to understand consequential assumptions and be able to challenge or recover the work. Oversight should match the task's scope and consequence rather than duplicate all its mechanics.
Does iatroX provide Heidi II certification or guaranteed protection against deskilling?
No, the learning methods described here are not Heidi II certification or a proven antidote to deskilling. Their educational value and the competence required for a particular clinical system need appropriate assessment.