Who Owns the Mistake? The Career Hiding Inside AI’s Hardest Question
Written by Rod on September 7, 2026
Series: From Bedside to Build · Episode 10 · Pillar 3: Expertise and Opportunity
By Rod Gamble | Week 37, 2026 | Pillar 3: Expertise and Opportunity
Years ago I broke ICU documentation across thirty hospitals in one afternoon — and what saved me wasn’t technical skill at all.
I’ll tell that story properly in a minute, because I have never needed it more than I do this September. If you work anywhere near a health system right now, you are living through the season when artificial intelligence stops being a pilot project and starts being the furniture — and one question is travelling through every break room ahead of it.
The Season the Pilots End
By the time you read this, the NHS will be weeks away from completing the largest AI deployment in its history — Microsoft 365 Copilot on the desktops of 505,000 clinicians and staff by October. In July, NHS England followed with an accelerated national AI rollout aimed squarely at waiting lists, framing the technology not as an experiment but as infrastructure. And at NHS ConfedExpo this summer, senior digital leaders called time on “pilotitis” altogether — NHS England’s national clinical lead for AI described continuing to pilot ambient voice technology as, his word, “nuts.” The message from the top of healthcare technology could not be clearer: the experimenting is over, the scaling has begun, and similar waves are moving through US and Irish systems.
But walk from the podium to the ward and the mood changes. When 233 registered nurses were surveyed about AI clinical decision support, 69.8% agreed that automated prioritisation would be useful. And 77.9% — read that number again — were concerned about who would be responsible if the AI made the wrong recommendation. Both findings are true at once. Useful, and unowned.
Therefore the question of the season isn’t whether the tools work. It’s the one clinicians keep asking quietly, and nobody on the slide deck seems assigned to answer: when this gets it wrong — and one day, somewhere, it will — who owns the mistake? Most clinicians treat that question as the reason to keep their distance from the whole subject. I want to convince you it is exactly the opposite. First, the story.
The Day I Took Down Thirty ICUs
In the late 2000s I was a Cerner clinical documentation consultant at one of the largest health systems in the American South. By then I was the person who had cracked NICU documentation so it finally made sense to nurses, and I’d solved ECMO documentation when almost nobody else in the country had — which, honestly, had made me a little cocky. One afternoon I needed to make a minor change, and I skipped my own hard-won discipline of testing it in the non-production system first. I made it straight in production.
It deleted vast amounts of configuration and took down ICU documentation across roughly thirty hospitals for about forty-five minutes. The moment I realised, my heart sank — I knew it was bad, and I knew it was avoidable. But here is the part that matters. The first thing I did was phone the organisation’s customer service line and tell them exactly what I had done, how I was fixing it, how long it would take, and that the calls were about to spike. I was right on every count, and I had the automated tools to rebuild fast. The complaints still came — but nobody had to waste hours investigating a mystery outage, because I had already told them what it was.
Two disciplines have travelled with me ever since. Keep your rigour exactly when your expertise tempts you to skip it — test before you touch production, every single time. And when something does break, owning it instantly and narrating the fix turns a crisis into a managed incident. But notice the deeper thing: the technology didn’t decide what that afternoon became. A person did. The system wasn’t safe again when the configuration was restored. It was safe the moment someone credible said — this is what happened, this is what I’m doing about it, and this is when it ends.
The Question Is a Job Description
Here is the reframe I most want you to take from this piece. When clinicians ask who is responsible when the AI gets it wrong, they assume the answer has to come from somewhere else — the vendor, the IT department, the board. But accountability in clinical systems has never really lived with the technicians. It lives with the people who understand what a wrong recommendation costs at the bedside, and who can design the workflow that catches it before it lands. That is not technical work wearing a clinical costume. It is clinical work — and almost nobody is formally doing it yet.
Look honestly at the evidence and you can see the shape of the job. A quality improvement study of 263 physicians and advanced practice providers across six US health systems found ambulatory burnout fell from 51.9% to 38.8% after just thirty days with an ambient AI scribe — a real, humane result. But a 2026 systematic review of AI supporting nurses’ clinical decision-making was markedly more equivocal about outcomes across settings. Both can be true, because these tools help unevenly — brilliantly in one workflow, clumsily in the next. Therefore every gap between the promise and the practice needs a clinician to find it, name it, and build the check that closes it. That is clinical innovation in its most practical form.
The market has already noticed. Roughly 70% of healthcare employers now treat digital competency as a critical hiring factor even for traditionally clinical roles, and the fastest-growing jobs are hybrids — informatics specialist, implementation lead, chief nursing information officer. Behind each scaled deployment sits the work almost nobody glamourises: safety cases, escalation routes, audits of what the model missed, the honest one-page account of where it fails. None of it requires a computer-science degree. All of it requires someone the room already trusts — someone who can hold the clinical consequence and the digital transformation in the same head. That person is not waiting to be trained somewhere. That person is reading this.
Steps You Can Take Now
Learn your incident route. Find out, this week, how an AI-related error would actually be reported where you work — and who holds clinical safety responsibility for the tools you use. In the UK, health IT deployments are required to have a named Clinical Safety Officer; most clinicians have never asked who theirs is. Asking is the first act of ownership.
Keep a failure log. For the one AI tool in front of you, note where it saves time, where it gets things wrong, and what would have caught the error. A month of honest entries is governance experience — evidence of exactly the judgement these programmes are missing.
Ask the accountability question as design, not protest. In the next meeting where the new tool comes up, ask it plainly: “When this is wrong, what happens — and who owns it?” Then offer to help draft the answer. The person who asks in order to build is remembered very differently from the person who asks in order to resist.
Put your hand up when the governance group forms. AI safety committees, super-user networks and workflow review boards are being stood up across health systems right now, and they are chronically short of working clinicians. Nearly every digital health career I know — including mine — began with a hand going up.
The machines were always going to make mistakes. So do we — that was never the disqualifier. What makes a system safe is that someone credible owns the error, narrates the fix, and redesigns the workflow so it doesn’t land twice. AI doesn’t remove responsibility from healthcare. It concentrates it — and the clinicians willing to carry that weight are the ones who will decide what these tools become.
When you’re ready to talk, rodgamble.com is where to find me.
References
1. NHS England. “500,000 NHS staff to get new artificial intelligence tools to help free up more time for patients” (June 2026). https://www.england.nhs.uk/2026/06/500000-nhs-staff-to-get-new-artificial-intelligence-tools-to-help-free-up-more-time-for-patients/
2. NHS England. “NHS accelerates artificial intelligence rollout to cut waiting times and improve care for millions” (July 2026). https://www.england.nhs.uk/2026/07/nhs-accelerates-artificial-intelligence-rollout-to-cut-waiting-times-and-improve-care-for-millions/
3. Digital Health. “NHS leaders call time on AI pilots and demand national scaling” (June 2026). https://www.digitalhealth.net/2026/06/nhs-leaders-call-time-on-ai-pilots-and-demand-national-scaling/
4. “The Influence of Artificial Intelligence on Clinical Decision Making: Geriatric Registered Nurses’ Perspectives” (survey of 233 RNs). PubMed Central. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC10736858/
5. “Use of Ambient AI Scribes to Reduce Administrative Burden and Professional Burnout.” JAMA Network Open (quality improvement study, 263 clinicians, six health systems). https://pmc.ncbi.nlm.nih.gov/articles/PMC12492056/
6. Mikkonen et al. “Artificial Intelligence Technologies Supporting Nurses’ Clinical Decision-Making: A Systematic Review.” Journal of Clinical Nursing (2026). https://onlinelibrary.wiley.com/doi/10.1111/jocn.70156
7. INNOVA People. “How digital health is changing healthcare careers” (May 2026). http://innovapeople.com/2026/05/25/how-digital-health-is-changing-healthcare-careers/
