What we’re learning from healthcare leaders, client conversations, and the changing realities of AI at work.
AI is moving quickly across healthcare. Ambient documentation, copilots, and automations that employees build themselves are already part of everyday work. But the harder part isn’t adopting the technology. It’s preparing people to use it well.
In our work with healthcare organizations, and in conversations with leaders across the industry, the same questions keep coming up. How do you build confidence without creating overconfidence? What does the workforce actually need to understand about AI? How do you encourage experimentation and still manage risk? And how do you plan for readiness when the technology keeps changing? Five themes are emerging.
1. The best AI opportunities start with the work, not the technology.
Many of healthcare’s most promising AI applications solve a problem people already feel. Ambient documentation is a good example. For some clinicians, spending less time on documentation means more time with patients. But the value varies a lot by role, specialty, setting, and workflow. A tool designed around how physicians work may not translate to nursing, and something that saves one team hours can add friction for another.
So, AI strategy should start with the work itself. Instead of asking where they can use AI, organizations may get further by asking where work is unnecessarily hard today. Where are people losing hours to repetitive tasks? Where is documentation getting in the way of patient care? Where are staff manually gathering, summarizing, or moving information? The strongest use cases are the ones where the value is obvious to the people doing the work.
2. AI adoption is as much a people challenge as a technology challenge.
Formal education matters, but people also learn by watching colleagues they trust apply new technology to real work. We’re seeing organizations try physician champions, super users, communities of practice, office hours, hackathons, and prompt-a-thons so employees can learn from each other.
That points to two forces working together. Formal learning builds the foundation: a shared understanding of privacy, risk, approved tools, responsible use, and what the organization expects. Peer evidence creates the pull. Knowing the rules doesn’t make anyone want to change how they work. Hearing a coworker say “this used to take me three hours, now it takes 20 minutes” often does.
Readiness, then, isn’t just about more training. It’s about building an environment where people learn from the organization, from experts, and from one another.
3. AI literacy needs to build judgment, not just technical skills.
Prompting is useful, but it’s only part of using AI responsibly. People also need to understand privacy, appropriate use, human accountability, probabilistic outputs, hallucinations, and verification. They need to know when a response should be questioned or escalated.
The tricky part is that AI can be wrong and still sound convincing. A fluent answer isn’t necessarily an accurate one.
People have a version of this problem too. We all have knowledge gaps, and sometimes we hold incorrect information with high confidence. At Amplifire, we call this Confidently Held Misinformation™ (CHM). It matters because someone who knows they’re unsure is more likely to stop, check, or ask for help. Someone who is confident and wrong is more likely to act. AI adds another layer: now employees have to evaluate both what they know and what the AI appears to know.
That makes judgment a critical skill. AI literacy isn’t just getting an answer out of AI. It’s knowing what to do with the answer.
4. Governance has to match the risk and still leave room for experimentation.
“Keep a human in the loop” is a good principle, but as AI shows up in more workflows, organizations will need to be more specific about what meaningful oversight looks like. Not every use case carries the same risk. Summarizing an internal administrative report is very different from informing a clinical decision, and validation, monitoring, and accountability should reflect that.
Meanwhile, a lot of AI innovation is happening outside IT. Employees are building their own automations, and some of them save real time. That raises a practical question: when does one person’s experiment become an enterprise capability?
Organizations need a path that allows useful experimentation without assuming every experiment is ready to scale. One way to think about it: Experiment → Validate → Govern → Scale → Own. The goal isn’t unlimited experimentation, or governance so tight that good ideas stall. It’s a responsible path between the two.
5. AI readiness is exposing deeper organizational challenges.
AI makes work faster, and it also amplifies whatever is already there. Automate a bad process and you get a faster bad process. Give AI access to outdated documents and outdated information gets easier to find. Fragmented institutional knowledge becomes a bigger problem when one system can reach all of it. That’s why AI readiness keeps running into process improvement, data stewardship, knowledge management, and workforce development.
It also changes how organizations should think about education. Tool-specific training will still matter, but tools change fast. The most durable AI education focuses on capabilities that last, like judgment, appropriate trust, verification, privacy, and accountability. From there, organizations can layer in role-specific education and ongoing guidance for the tools and policies that change more often.
And the workforce isn’t starting from the same place. Some people are just beginning to try generative AI. Others are already building sophisticated automations. One course for everyone probably won’t serve either group well. AI workforce development will need to get more personal, not just more comprehensive.
What healthcare leaders should be asking now
There’s no finished playbook for this. The technology is moving too fast, and organizations are learning as they deploy. But the questions are getting clearer:
- Where is AI creating measurable value for our workforce today?
- What does every employee need to understand to make sound decisions with AI?
- How should human oversight change based on the risk of the use case?
- Are our processes, data, and institutional knowledge ready to be amplified by AI?
- How will we know people can apply what they’ve learned, not just that they completed training?
That last one may matter most. Licenses, adoption rates, training completions, and deployed use cases show how much AI is being used. They can’t really show whether people know when to trust it, when to question it, and when they might be confidently wrong themselves. Those are human questions, and they’ll only get more important as AI gets more capable.