Change management: GenAI and CDS are already accepted by clinicians

Most health IT leaders know the picture. Everett Rogers drew it in 1962. A bell curve. Innovators at the left, then early adopters, early and late majority, laggards at the right. Technology diffuses slowly, over years. Resistance is expected. Change is managed.

Rogers was right about almost everything in digital health. His model is not right for generative AI.

Introduced commercially in November 2022, GenAI quickly produced three new clinical software categories. By the end of 2025, ambient scribes had moved from novelty to tipping point: a CMA-CFIB survey found 28 percent adoption among Canadian physicians.

OntarioMD and Infoway created safe harbour VORs. No enterprise implementation. No steering committee. Few pilots. Clinicians downloaded an app, often before their IT department knew. Evaluations showed remarkably positive results.

Second-screen clinical decision support followed a faster pattern with even less governance. Tools like OpenEvidence and Doximity sit beside the EMR, answer clinical questions in real-time, touch no PHI, and spread through peer recommendation.

OpenEvidence reported 34 percent of Canadian physicians using its tool as of February 2026. Clinicians adopt these tools the way they adopt journal subscriptions, not the way they adopt EMR modules. OE and Dox have been joined by Heidi, Tali and others.

On the patient side, AI usage is skyrocketing. Some studies showing a third or more of web searches now converting to GenAI. Patients bring their own GenAI tools into the exam room to record and transcribe the encounter in real-time. The contrast with portal adoption and frustration could not be more stark.

This does not happen in digital health. It is happening in GenAI. The two are not the same thing.

Wachter’s Reversal: Robert Wachter’s 2015 book The Digital Doctor is a deeply ambivalent account of what the EHR did to the medical profession. The documentation burden exploded. Technology turned clinicians into expensive data-entry clerks.

Wachter’s new book, A Giant Leap (2026), marks a striking reversal. GenAI is not an extension of the EHR era but its corrective. The ambient scribe is his central exhibit: a technology solution to a problem that technology created. The scribe restores eye contact. It does not fix the EHR. It sits above it, and it works well:

“The speed with which the digital scribe went from not-ready-for-prime-time to something every doctor wanted offers several insights about both the implementation and business considerations surrounding AI in healthcare.” (Wachter, A Giant Leap, p. 97)

GenAI is empowerment, not procurement. It is a knowledge strategy, not a release strategy. Leaders who mistake it for another digital health project will specify old solutions, lock into old technology, and find their clinicians and patients have already moved on.

Taylor and Ford Versus Deming and Toyota: Digital health is a Taylor and Ford technology. Frederick Taylor decomposed work into discrete tasks and optimized each one. Henry Ford built the assembly line. Together they gave the auto industry (and health IT) a model that treats the existing process as fixed and asks how technology can make it faster and more legible.

The EHR digitized the paper chart. CPOE digitized the medication order. The process was not questioned. It was encoded.

Generative AI is closer to Deming and Toyota. W. Edwards Deming argued that quality is built in, not inspected out, and that the people closest to the process know where it fails. Toyota built a production system on that principle. The tool does not impose a fixed process. It responds to the one in front of it.

Digital health asked: how do we digitize what we already do? Generative AI asks: given an intelligence layer that can do work for us, what should we do instead? That question changes scope of practice, models of care, administrative need, and clinical communication. Those are different change management problems.

Managing change in the late twenties: The change management challenge in generative AI is not resistance. Resistance was the digital health problem. Leaders built entire disciplines around overcoming it: communication plans, physician champions, super-user networks.

Those tools sound quaint when a third of clinicians are already using GenAI. By the time this is published it will be more than half. By the end of 2027 the vast majority of Canadian physicians will be using GenAI. Probably 80+ in the authors’ opinions.

Adoption is outrunning governance. The existing models that conflate “clinician engagement” with months of committee process are not merely inefficient. They are a risk. Exhausted clinicians will not tolerate project plans that prioritize process over relief.

The era of multi-month design phases followed by rigid activation and sustainment windows is over. Those timelines belonged to the EMR era. The EMR era ended sometime in 2025.

GenAI demands a shift from perfection to iteration. Change management is not imposed from the top down. It is pulled from the bottom up. Organizations do not need to overcome resistance. They need the agile capacity to govern a tool that clinicians already use and patients already demand.

The ambient scribe and second-screen CDS give time back. Five to 15 hours a week of pyjama time, returned. What health systems do with that time is the new change management question. It is a Deming quality improvement problem.

GenAI will move faster still: Everything described above is already being overtaken. Vibe coding builds functional software through natural language prompts and compresses the gap between idea and deployed tool from months to days.

A clinician with a problem and access to an AI coding agent can prototype a solution before a procurement committee drafts its terms of reference. Autonomous agents take multi-step actions, retrieve information, complete forms, coordinate handoffs, and communicate with patients.

Meanwhile, scribes and CDS are converging. Both are adding referrals, note templates, history-taking, and a widening range of clinical functions. The next generation of clinical copilots will be difficult to distinguish from one another.

Every major EMR vendor acknowledged these emerging facts in late 2025 by announcing forthcoming scribes, CDS tools, agents, and copilots. At least eight major technology companies have entered their space. Legacy integration was an advantage in digital health. It may be a disadvantage when the intelligence layer arrives.

In the spring of 2026, it is an open question whether the EMR retains its position at the top of the clinical technology stack or becomes a system of record: necessary infrastructure, but no longer the organizing layer of clinical work. A system of intelligence may become the interaction layer that adapts to the clinician, replacing the legacy method of changing the clinician to adapt to the EMR.

Jennifer MacGregor is Managing Director of Elewana Advisory; Will Falk is a policy fellow at the CSA Public Policy Group, the C.D. Howe Institute, Rotman, and WiHV.