The Computer Knew the Rule. It Didn’t Know the Room.
Series: From Bedside to Build · Episode 2 · Pillar 3: Expertise and Opportunity
By Rod Gamble | Week 29, 2026 | Pillar 3: Expertise and Opportunity
Years ago I found a safety rule buried in a hospital’s patient-record system that was protecting almost everyone in the building — and quietly endangering the sickest patients in it.
The rule was simple and sensible. Before a nurse could give a medication, a pharmacist had to review the order first. In almost every ward of that hospital, prospective pharmacy review was exactly the safeguard you would want. But in two places — the Emergency Department and the Post-Anaesthesia Care Unit — the gap between a physician saying “give this now” and the drug actually reaching the patient had to be as short as humanly possible. There, the same rule that kept everyone else safe became a barrier to care. A clinical hazard, dressed up as a safety feature.
The software could not see the difference. It knew the rule. It did not know the room.
So I went and built the difference. I pulled in an expert in the pharmacy module and an expert in the system’s programming language, and together we wrote a solution that recognised when a user was physically standing in the PACU and, only there, lifted the requirement for review before administration. Everywhere else, the safeguard stayed exactly as it was. Care in the place that needed speed got faster and safer. Care everywhere else stayed protected. One rule, made intelligent about context.
I have been thinking about that fix a great deal in 2026, because the whole of healthcare is now living a much larger version of it.
2026 Is the Year the Bill Came Due
For two years, clinical AI arrived faster than anyone could govern it. Tools that draft your notes, summarise a chart, surface a care gap, suggest an order. The market reflects the stampede: AI in clinical workflow was valued at around $2.8 billion in 2025 and is forecast to pass $45 billion by 2035, growing at better than 30% a year. Clinicians, exhausted and under-resourced, did not wait for permission. They reached for whatever saved them an hour. By 2025 that had a name — shadow AI: staff quietly pasting patient information into consumer chatbots that were never approved, never secured, never reviewed.
Now the bill has come due. The people who track this closely are calling 2026 “the year of governance” — the year hospital leadership finally has to catch up to the clinicians who already adopted the technology. Written AI policies. Approved use cases. And the part that matters most for you: a hard requirement that a clinician validates the AI’s output before it touches the record.
Read that again, because it is the whole opportunity. The model can produce the first draft. It cannot be trusted to know the room. Someone with clinical judgement has to stand between the output and the patient. That someone is not a data scientist. It is a clinician.
The Asset You Built at the Bedside
Here is what I want every burned-out nurse, doctor, and allied health professional to understand. The thing that makes AI safe in healthcare is exactly the thing you spent years acquiring and now undervalue: the ability to know that a rule which is right in one room is wrong in the next.
An algorithm can learn the rule. It cannot feel the difference between a routine med order on a quiet ward and the same order called out across a PACU bay with the patient in front of you. That difference — context, judgement, the instinct earned over thousands of shifts — is the scarcest input in the entire AI supply chain. And it is the one thing the technology cannot manufacture.
This is why the digital health workforce keeps growing while everyone else worries about being automated away. Health informatics roles in the US are projected to grow about 16% through 2033 — far faster than average — and the clinicians moving into them report some of the highest career satisfaction in the field, with the majority working remotely at least part of the week. The demand is not for people who can build the models. It is for people who can govern them. People who have walked the ward.
That is the move from bedside to build. You are not abandoning your clinical expertise. You are repricing it — from something rented by the hour at the bedside into something that shapes how care is delivered for everyone. This is the heart of clinical transformation: digital transformation does not replace clinical innovation, it depends on it.
Steps You Can Take Now
You do not need to learn to code to start. You need to start treating the frictions you already see as the asset they are.
First, keep a friction log. For two weeks, write down every moment a system, a rule, or a workflow fights the care you are trying to give — the alert that fires at the wrong time, the field that does not fit your reality, the AI suggestion that was confidently wrong. This is the raw material of clinical informatics. You are already collecting it in your head; start writing it down.
Second, learn the language of governance. Read your organisation’s emerging AI policy. Find out who sits on the AI or clinical informatics committee. The vocabulary — validation, expert-in-the-loop, post-market monitoring — is learnable in a few weeks, and it is the bridge between your clinical voice and the people making the build decisions.
Third, volunteer your judgement where it is now being asked for. Every health system standing up AI governance needs clinicians to review outputs, define safe use cases, and flag the rooms where a sensible rule becomes a hazard. Put your hand up for that work. It is the most direct on-ramp from clinician to clinical informaticist I know, and it pays in both credibility and options.
Start there, and within a few months you will have something most clinicians never build: documented evidence that you can see what the computer cannot.
The computer knew the rule. It took someone who had stood in the room to know the rule was wrong. In 2026, healthcare is quietly admitting it needs far more of those people — and you may already be one of them.
When you’re ready to talk, rodgamble.com is where to find me.
References
1. Wolters Kluwer. “2026 healthcare AI trends: Insights from experts.” https://www.wolterskluwer.com/en/expert-insights/2026-healthcare-ai-trends-insights-from-experts
2. Wolters Kluwer. “Shadow AI: A hidden risk to healthcare.” https://www.wolterskluwer.com/en/solutions/uptodate/ai-clinical-decision-support/shadow-ai-report
3. Foley & Lardner LLP. “Aaron Maguregui Shares Insights on Shadow AI Risks in Health Care” (March 2026). https://www.foley.com/news/2026/03/maguregui-shared-insights-on-shadow-ai-in-health-care/
4. InsightAce Analytic. “AI in Clinical Workflow Market Size, Scope and Trends 2026 to 2035.” https://www.insightaceanalytic.com/report/ai-in-clinical-workflow-market-/3509
5. npj Digital Medicine (Nature). “Advancing healthcare AI governance through a comprehensive maturity model based on systematic review.” https://www.nature.com/articles/s41746-026-02418-7
6. Sciences (MDPI). “Governing Healthcare AI in the Real World: How Fairness, Transparency, and Human Oversight Can Coexist: A Narrative Review.” https://www.mdpi.com/2413-4155/8/2/36
7. Research.com. “Health Informatics Career: Guide to Jobs & Salary (2026).” https://research.com/careers/health-informatics-career-guide-to-jobs-and-salary
8. Nurse.org. “Top Health Informatics Careers 2026.” https://nurse.org/healthcare/health-informatics-careers/
