AI for Healthcare Workers — Not the Hype, the Real Application
I train healthcare workers on clinical systems for a living. Here's what AI actually does well in healthcare, what it doesn't, and what the people using it day-to-day actually need.
The AI discourse in healthcare is mostly written by people who've never watched a floor nurse try to use a new clinical system for the first time. The actual experience of healthcare technology at the point of care is different from the conference deck version.
I need to be clear upfront: I train healthcare workers on clinical software. Nothing in this post touches patient data, clinical specifics, or anything PHI-adjacent. What I'm drawing on is years of watching real people — nurses, techs, coordinators, administrators — interact with technology systems in a clinical environment.
The AI discourse in healthcare tends to come from two places: vendors selling AI-powered clinical tools, and researchers studying AI's diagnostic accuracy. Both groups are largely missing the actual human experience at the point of care.
Here's what that experience actually looks like.
what healthcare workers are dealing with
Healthcare workers are already drowning in technology. The EHR alone is a system of enormous complexity that requires significant training, produces high cognitive load, and interrupts clinical workflows in ways that are frequently at odds with what the worker actually needs to do.
Adding AI to that environment doesn't feel like adding a helpful assistant. It often feels like adding another system to learn, another screen to check, another thing that might be wrong in a way that matters.
The healthcare workers who are most skeptical of AI tools aren't skeptical because they're technophobic. They're skeptical because they've seen technology promises not match reality, and because the cost of a tool failure in their environment is different from the cost of a tool failure in a software company.
what AI actually does well in clinical workflows
Documentation assistance is the clearest win. The amount of time healthcare workers spend on documentation is enormous and genuinely reduces time with patients. AI that can listen, structure, and draft documentation — with the clinician reviewing and approving — reduces that burden meaningfully.
Prior authorization and insurance navigation. This is tedious, rules-based, and time-consuming. It's also exactly the kind of structured work that AI handles well. Reducing the time a nurse or coordinator spends on prior auth is an unambiguous quality-of-care improvement.
Clinical decision support — not diagnosis, support. AI that surfaces relevant protocols, flags potential drug interactions, or prompts for missing data elements at the right moment in a workflow is genuinely useful. AI that tries to diagnose is entering territory where the stakes are too high for current reliability.
Training and knowledge support. This is where I work. Healthcare workers need to learn large, complex systems while also doing their actual job. AI that can answer "how do I do X in this system" in plain language, in the moment, is valuable in a way that static documentation never was.
what AI doesn't do well there
Interrupting time-critical workflows with prompts that require decision-making. If a clinician is in the middle of a critical task and an AI system interrupts with a suggestion that requires evaluation, you've added cognitive load to the worst possible moment.
Replacing judgment calls that require full situational context. The AI doesn't know that the patient is also dealing with Y, that the family has raised Z, that the clinical picture has shifted since the last data update. Healthcare AI that presents recommendations as if it has complete situational awareness is dangerous.
Systems that require explanation when they're wrong. A clinician who trusts an AI recommendation and acts on it has to be able to explain that action. "The AI suggested it" is not a defensible clinical decision. The technology has to be integrated into workflows in ways that preserve human accountability.
what healthcare workers actually need from AI
Simple. Fast. Interruptible and resumable.
A tool that requires a 2-minute workflow to get a 30-second answer is worse than no tool. A tool that can't be abandoned midway through because the patient is calling is worse than no tool.
Transparency about confidence. Not in technical terms — in plain language. "This is a common pattern in your documentation" is different from "Based on the ICD-10 mapping algorithm, there is a 0.78 correlation..." Healthcare workers need to be able to assess whether to trust the output without needing a statistics background.
Integration that doesn't add screens. The best clinical AI is invisible until it's useful — surfacing at the right moment in the existing workflow, not requiring the worker to go somewhere else.
from my own bench
The moments where I've seen technology genuinely make healthcare workers' lives better are the ones where someone thought hard about the actual workflow, the actual pressure, and the actual cognitive load — and designed the tool to fit inside those constraints rather than asking the worker to adapt to the tool.
That design work is hard. It requires spending time with people doing the job, not just the people buying the technology. Most healthcare AI has more investment in the capability than in that design work.
The workers who do well with new clinical technology are the ones with good foundational habits: they learn the tool deliberately, they know what to trust and what to verify, and they don't let the technology make decisions they should be making. Those habits apply to AI too.
the bottom line
Healthcare AI is real and some of it is genuinely useful. The gap between the vendor deck and the experience at the bedside is significant. The tools that will actually improve care are the ones designed around how clinical work actually flows — not how it looks on paper.
That design work is the hard part, and it's where most of the opportunity is.
— Dru Edwards