Series: From Bedside to Build · Episode 14 · Pillar 3: Expertise and Opportunity
By Rod Gamble | Week 41, 2026 | Pillar 3: Expertise and Opportunity
For twenty years I have sat in the rooms where your software gets decided, and I can tell you how few clinicians are in them.
That is not a boast. It is closer to an admission. Programme manager on a green-field electronic record build across two NHS Trusts. Senior vice president on a global service line. Programme director on a six-country hospital system. In most of those rooms, the number of people who had recently done the work the software was about to change was zero, or one — and the one was usually me.
The Argument Has Moved
For three years the debate about AI in healthcare was existential. Will it replace me. Can a model do what I do at the bedside. That argument is essentially over, and it ended in a draw nobody finds very interesting.
Listen instead to what clinicians are saying now, in public, where they assume no vendor is reading. A physician wrote on a doctors’ forum in May that where ambient AI is only marginally useful, the value from using it “is not accruing to you as the clinician but to another layer of admin.” A nurse, on a nursing forum: technology is being pushed into shifts “without anyone asking us if it actually helps.”
Neither is a complaint about technology. Both are complaints about arithmetic — about who ends up holding the saving.
And there is now a real number to hold. A JAMA study published this spring, covering 8,581 ambulatory clinicians across five academic health systems, found that AI ambient scribes were associated with 13.4 fewer minutes of total electronic health record time and 16.0 fewer minutes of documentation time — both measured per eight scheduled patient hours. Real, measurable, and modest. The senior author’s own word for the reductions was “modest.”
So here is the question the slide decks skip. Thirteen minutes exist. Who gets them?
Nobody promised you those minutes. They only promised the minutes existed. In the same study, clinicians using the tool saw 0.49 more visits per week — and time spent in the record outside scheduled hours, the thing clinicians actually call pajama time, did not change significantly at all. The saved minutes did not come home with anyone. They went back into throughput, for an average of $167.37 a month in additional billing per clinician.
The Room You Were Not In
Therefore the honest frame here is not adoption. It is authorship. Somebody decided what those minutes were for, and it was not the person who saved them.
We now have numbers on how few clinicians are in that decision. Elsevier’s Clinician of the Future 2026: Nurses Edition surveyed 692 nurses and 2,065 doctors across 118 countries. 40% of nurses said they have no seat at the table where AI decisions are made; only 42% said the tools are trustworthy today. The National Nurses United survey of more than 2,300 registered nurses found 60% did not trust their employers to prioritise patient safety when implementing AI, and 69% said AI-driven acuity tools did not match the care their patients actually needed.
Sit with that last figure. Roughly two-thirds of the people doing the assessment say the score disagrees with the assessment. Somebody wrote that algorithm — chose its inputs, its thresholds, its escalation rule. That somebody had a job title, a salary and a seat in a room, and in most builds I have worked on, no NMC or GMC registration number.
Which is why “learn AI so you stay employable” lands so badly right now. It is very nearly the sentence the people running the rollout are using, and clinicians have already noticed. Adaptation is not the offer. Adaptation is the thing being done to you.
A Nurse Redesigning the Process
But I have seen the other arrangement, and it is worth describing, because it is not exotic. A couple of years after I left bedside nursing, I spent a year as an automation engineer at Intel, in Rio Rancho. My job was a process used worldwide for end-of-wafer handling. It scaled badly — the workload grew with the square of the input, so every rise in volume cost far more than it should have. I redesigned it to grow linearly instead, and along the way wrote down how large parts of it actually worked, because nobody ever had.
The point is not “a nurse can learn to code,” though that happens to be true and I am reasonable evidence for it. The point is what that company did that healthcare currently does not. They put the redesign of the work in the hands of somebody whose job was to understand the work. In manufacturing that is unremarkable. In healthcare in 2026 it is nearly unheard of — we are redesigning clinical workflow at national scale and staffing the design with people who have never done the workflow.
The Work That Cannot Be Outsourced
But here is the turn, and it is better news than it first sounds. The gap will not be closed by clinicians learning to operate tools. It will be closed by clinicians checking them — and that work has a floor beneath it no vendor can dig under.
In England there are two clinical safety standards, and the distinction matters. DCB0129 governs the organisation that manufactures a health IT system. DCB0160 governs the organisation that deploys and uses one — your trust, your practice, your ICB. Both require a named Clinical Safety Officer, and that officer must be a registered healthcare professional with appropriate clinical safety training. Not a data scientist. Not a product manager. A registered clinician, whose name goes on the clinical safety case report. Meanwhile a national cross-sectional study of digital health technologies in NHS England found that only 17.3% were compliant with those clinical safety standards. The training pipeline for the role is narrow, and demand outstrips supply.
The same signal is everywhere. The American Nurses Association spent this year calling publicly for nurse-led guardrails on AI. In June, three London health innovation networks published a report arguing the NHS conversation has to move past productivity toward governance, permission to use, and workforce development. That is the institutional way of saying oversight is now real work, and it is unstaffed.
The arithmetic follows the scarcity, as it usually does. Nearly 60% of nurse informaticists now earn more than $100,000 a year, up from 49% in 2020; among ANCC-certified informatics nurses it is 85%. But I am not offering that as an escape hatch, and I would rather you distrusted me than believed I was. The market for informatics qualifications is crowded, the jobs cluster geographically, and the degree counts for little without clinical experience behind it. Almost everyone I know in one of these roles did not walk through a door — they expanded a role they already held until it quietly became a different job. Nobody hands you the room. Nobody can staff it without you either.
Steps You Can Take Now
All of this is done from where you stand, in hours you already work. None of it requires you to resign, retrain, or announce anything.
1. Keep a divergence log. For four weeks, note every time an AI output — an acuity score, a draft note, a suggested order, an automated handover — disagrees with your own assessment. Date, tool, what it said, what you assessed, what you did. Six lines a week, and almost nobody is writing it down.
2. Find out who owns the build. Every trust and health system has a digital clinical safety lead, a clinical informatics team, or an AI governance group. Find the actual name, and email them. Bring the log rather than an opinion.
3. Look up your system’s assurance route. In England, read DCB0160 — the standard for organisations deploying health IT — and the Clinical Safety Officer requirement inside it, then ask who holds that role where you work and what training your organisation funds. In the US, find your AI governance committee and check its clinical membership. In Ireland, ask your organisation’s digital or eHealth lead which clinical safety standard your deployments are assured against.
4. Say yes to the unglamorous committee — the superuser group, the optimisation working group, the order-set review. It looks like a detour. It is where clinical judgement gets written into the build.
5. Write one workflow down. Take a process you know cold and document how it truly runs, including the parts everyone works around. That is what I produced at Intel almost by accident, and it mattered more than the redesign did.
The Last Thing
Twenty years in those rooms taught me one thing worth passing on. The people deciding what happens to your thirteen minutes are not smarter than you, and they are not hiding from you. They are simply the ones who turned up holding a written record of how the work actually goes.
The rooms where this gets decided are not full. They are short of exactly one thing — and you have been carrying it the whole time.
When you’re ready to talk, rodgamble.com is where to find me.
References
1. Ahmad, H. et al. / JAMA Network. “Ambient AI Scribes and Clinician Electronic Health Record Time.” Reported by STAT News, 1 April 2026. https://www.statnews.com/2026/04/01/ai-ambient-scribes-modest-time-savings-clinical-documentation/
2. Mass General Brigham. “AI Scribes Linked to Modest Reductions in EHR Documentation Time.” https://www.massgeneralbrigham.org/en/about/newsroom/press-releases/ai-scribes-linked-to-modest-reductions-in-ehr-documentation-time
3. Elsevier. “Clinician of the Future 2026: Nurses Edition” (692 nurses, 2,065 doctors, 118 countries, Dec 2025–Feb 2026), reported by Nurse.org. https://nurse.org/news/nurses-ai-trust-elsevier-2026/
4. National Nurses United. “National Nurses United survey finds A.I. technology degrades and undermines patient safety.” https://www.nationalnursesunited.org/press/national-nurses-united-survey-finds-ai-technology-undermines-patient-safety
5. American Nurses Association. “American Nurses Association Calls for Nurse-Led Guardrails on Artificial Intelligence in Healthcare” (2026). https://www.nursingworld.org/news/news-releases/2026-news-releases/american-nurses-association-calls-for-nurse-led-guardrails-on-artificial-intelligence-in-healthcare/
6. Health Innovation Network South London, UCLPartners and Imperial College Health Partners. “New report calls for workforce-centred AI adoption” (June 2026). https://healthinnovationnetwork.com/news/new-report-calls-for-workforce-centred-ai-adoption/
7. NHS England. “Digital clinical safety assurance” (DCB0129 / DCB0160 and Clinical Safety Officer requirements). https://www.england.nhs.uk/long-read/digital-clinical-safety-assurance/
8. “Digital Health Technology Compliance With Clinical Safety Standards in the National Health Service in England: National Cross-Sectional Study.” https://www.ncbi.nlm.nih.gov/pmc/articles/PMC12619009/
9. HealthTech Magazine. “Preparing an AI-Ready Nursing Workforce: How Informatics Bridges Technology and Patient Care” (February 2026). https://healthtechmagazine.net/article/2026/02/preparing-ai-ready-nursing-workforce-how-informatics-bridges-technology-and-patient-care
10. U.S. Bureau of Labor Statistics. “Health Information Technologists and Medical Registrars” (Occupational Outlook Handbook). https://www.bls.gov/ooh/healthcare/health-information-technologists.htm
11. Student Doctor Network forums. “AI is Evil” (public forum thread, May 2026). https://forums.studentdoctor.net/threads/ai-is-evil.1519308/
12. allnurses.com. “Do you think AI will replace Nurses one day?” (public forum thread, August–October 2025). https://allnurses.com/you-think-ai-replace-nurses-one-t768714/
