The time goes wherever a practice's working arrangements allow it to go. Heidi Work could help a clinician finish on time, create space for complex care or support additional appointments, but those are different outcomes. A technical improvement does not decide how its benefits will be distributed.
Heidi's 22 September 2026 announcement describes a move into supervised action and says the coming capabilities will not be available in the UK and EU. The discussion here is prospective: assume substantial authorised task execution, then ask what would happen to the capacity released. It is not a report of measured Heidi Work savings.
The same saving can support different versions of a better day
A reduction in administrative handling can have several possible destinations. Work previously completed at home could fit within the scheduled day. A clinician could spend longer explaining a complex decision. An overdue backlog could become manageable. A practice could improve continuity by creating room for follow-up with the usual clinician. It could also add appointments.
None of these outcomes is automatically the correct one for every service. A practice with serious access problems may reasonably prioritise additional availability. A team struggling with sustained overtime may regard an ordinary finishing time as the most important improvement. A clinic with fragmented care may value continuity more than another isolated contact.
The difficulty arises when different people expect different benefits from the same implementation. The phrase "time saved" can conceal that disagreement until the appointment template changes.
For Heidi Work, a useful adoption question is therefore whether clinicians recognise the eventual working day as the one they thought they were helping to create. That is an incentive hypothesis to examine, not a prediction that any particular deployment will intensify work.
A fictional practice discovers two incompatible expectations
Imagine a practice considering delegated handling of agreed appointment changes and routine follow-up administration. Its clinicians support the idea because unfinished actions regularly extend their day. The practice manager supports it because patients struggle to obtain suitable appointments. Reception staff hope it will reduce repetitive calls.
During the proposed pilot, the agent appears to reduce some manual handling. Before the team has established how much work remains elsewhere, the organisation adds activity to the timetable. Clinicians now complete more consultations but still deal with exceptions after the session. Reception staff receive fewer simple requests but more complicated queries about actions already in progress.
This fictional scenario does not require an unreasonable manager or an ineffective agent. The problem is that three legitimate objectives were never reconciled. The original promise of relief and the later requirement for throughput are not identical propositions.
A different implementation could explicitly begin by reducing unfinished work, then review whether reliable spare capacity remains. Another could agree from the outset that the priority is access, while protecting time for exception handling. The important distinction is between an agreed arrangement and a benefit that is silently reallocated.
What existing GP research can and cannot tell us
The Nuffield Trust's research with the Royal College of General Practitioners, published on 3 December 2025, found that participating GPs described AI benefits through reduced administrative burden, less overtime and improved wellbeing. Their accounts included using recovered time for rest and self-care rather than automatically providing extra appointments.
That research concerns existing AI use and reported experiences, not an evaluation of Heidi Work or autonomous clinical agents. It cannot establish how much time a new product will release or how every clinician would prefer to use it. It does support asking clinicians directly what they regard as a meaningful benefit instead of assuming that productivity means more activity.
The distinction is commercially relevant. A supplier could demonstrate faster completion of a narrow task while the people expected to adopt it experience little improvement in their overall day. Conversely, a modest reduction in unfinished work could matter substantially to a clinician even without an increase in appointment numbers. Both possibilities require measurement at the appropriate level.
Fewer tasks could leave a more demanding remainder
Suppose routine coordination becomes easier while unusual situations still reach the clinician. The remaining work might contain a greater concentration of ambiguity, emotionally difficult conversations and decisions requiring judgement. The administrative task count could fall while the perceived intensity of the working day stays unchanged or increases.
That is a plausible hypothesis, not an inevitable consequence of automation. The opposite is also possible: fewer interruptions and fewer unresolved actions might make difficult decisions easier to approach. The evaluation should distinguish these possibilities rather than choose the preferred story in advance.
A practical assessment would examine when exceptions arrive, how much context staff need to recover and whether there is protected capacity to handle them. A queue of exceptions delivered at the end of the session might be experienced differently from the same work presented with adequate context throughout the day.
The composition of the remaining work matters as much as its volume. Counting eliminated clicks would not answer that question.
Agree the productivity bargain before changing the timetable
A proposed productivity bargain should describe what a demonstrated gain will be used for and how the team will know whether that gain is real. It should also recognise implementation work: configuring processes, correcting mismatches, supporting colleagues and explaining changes to patients.
The agreement need not be elaborate. A practice could state that the initial objective is fewer unfinished administrative actions, with appointment capacity reviewed only after the workflow has stabilised. Another service might prioritise quicker access while explicitly measuring whether additional work has shifted to nurses or reception staff. Those are examples of local choices, not universal implementation rules.
The key is to include the people who absorb the consequences. A doctor's reduced handling time should not be counted as an unqualified organisational saving when another team member must spend longer reconstructing what happened. Similarly, a practice manager's coordination work should not disappear from the business case because it takes place outside the agent's interface.
A transparent agreement also makes disappointing results easier to interpret. If the main benefit is better completion rather than more capacity, the team can decide whether that still justifies the investment instead of pretending that every benefit is interchangeable.
Measure the day, not just the automated step
Task-level measurement remains useful. It can show whether an authorised action reaches its endpoint and how much staff intervention it requires. But a sustainable working-day assessment should also examine after-hours work, outstanding actions, interruptions, patient access, continuity and staff experience.
These outcomes should be checked again after appointment templates, staffing or expectations change. Otherwise an early pilot may describe a temporary arrangement that no longer exists by the time the service makes its renewal decision.
Attribution needs care. A quieter month, a change in staffing or a temporary effort to clear the backlog could alter the same measures. A comparison period or staggered implementation can help investigate alternative explanations, although neither automatically eliminates them. The purpose is to understand the contribution of the workflow, not to attach every improvement to the software.
For clinicians seeking less overtime, the relevant outcome is work that genuinely stops following them home. For organisations seeking access, it is additional appropriate capacity without hidden deterioration elsewhere. For teams seeking continuity or quality, it is whether the released attention reaches those goals. There is no single productivity verdict that answers all three scenarios.
Frequently asked questions
Will time saved by Heidi Work automatically become more appointments?
No, additional appointments would be an organisational choice rather than an automatic consequence of faster task handling. The team should distinguish that objective from reducing overtime, improving continuity or creating more time for complex care.
Could automation make the remaining workload feel harder?
It could concentrate exceptions and difficult decisions, but it could also reduce interruptions and unfinished work. Both possibilities should be investigated through staff experience and workflow measurement rather than assumed in advance.
What should a practice agree before a pilot?
It should agree the intended use of any demonstrated capacity gain, the work included in the assessment and how effects on the whole team will be reviewed. Changes to appointment templates should be evaluated against those agreed objectives rather than treated as proof of success by themselves.
Read more iatroX analysis of clinical AI and professional practice →