AI in healthcare (take 2) — what’s now, and what’s next

When we wrote about AI in healthcare almost exactly two years ago, much of it lived in the realm of prediction. Some bets were safe, some were bold. Now we have the benefit of hindsight… and a new wave of capability pushing even further than we imagined.

Clinical Documentation: From Prediction to Practice

In the original post, we said:

“Most EHRs are ‘systems of records and storage’ with limited capabilities. Machine learning can transform EHRs from ‘systems of records’ to ‘systems of intelligence’.”

That wasn’t a throwaway line. We pointed to the possibility of voice-driven systems that could listen in, transcribe, and summarize visits so clinicians could spend less time buried in keyboards. Today that’s happening. Epic, Nuance, and Abridge all have ambient AI scribes in production across major health systems. Early reports show substantial reductions in after-hours charting and rising clinician satisfaction. What was hypothetical two years ago is now part of everyday workflows.

Medical Imaging: Evidence Over Hype

Back then we wrote:

“Most experts don’t think AI will replace radiologists, but rather extend their capabilities by doing first glance analysis for radiologists.” 

That sounded like a helpful assist. The reality has gone further. Prospective studies in mammography show AI-supported reads are improving cancer detection rates without increasing false positives. Colonoscopy tools are consistently boosting adenoma detection. It’s no longer just about efficiency — the evidence suggests real improvements in clinical outcomes.

Patient Engagement: From Chatbots to Care Managers

Our earlier take was modest:

“You’ll also almost assuredly start to see chatbots within EHRs so that patients can ask general questions about their treatment plan and get immediate answers, powered by robust data.”

That baseline has been overtaken by what’s actually rolling out. One really cool example: Sage, from Ellipsis Health, is a fully autonomous care manager that calls patients, checks eligibility, manages enrollment, and follows up on complex cases across physical, behavioral, and social health needs. Ellipsis reports their clients (healthcare systems, doctors’ offices, etc.) are reporting a 60% reduction in administrative work, sixfold faster enrollment, and measurable ROI. This isn’t a reminder bot — it’s an empathetic, AI-driven care manager that reshapes the patient touchpoint itself.

Personalized Treatments: Beyond Boilerplate Medicine

From our original article:

“AI could crunch massive amounts of data to determine not only optimal treatment courses, but tailor that to the information gleaned from your genomic testing… your care could truly be personalized based on your genes + insight from AI datasets.”

That personalization layer is maturing (and it now taps new structural biology tools). Enter AlphaFold’s step-change (see next section) and a broader wave of multimodal models that combine text, imaging, labs, and omics for triage and treatment selection. On the bench-to-bedside path, the most concrete advances today are still in imaging + workflow (above) and documentation (above), while the deeper biology piece is getting a powerful upgrade.

AlphaFold: The “Someday” That Already Happened

We closed the original piece with this prediction:

“AlphaFold could be a Nobel-Prize winning development some day…”

Man did we NAIL that one. That “someday” came last year — the team behind AlphaFold won the Nobel Prize in Chemistry for transforming protein structure prediction and computational protein design. The recognition signals that AI models aren’t just tools; they’re advancing basic science to the point of reshaping medicine itself. 

With AlphaFold 3 now modeling protein-ligand and protein-DNA interactions, the path forward for drug discovery is even more expansive.

The Newer Wave

So what’s next? The frontier isn’t just doing old tasks faster, but combining modalities — text, imaging, genomics, and even patient voices — into a unified view of health. Imagine ambient scribes feeding structured data into predictive models, imaging AI cross-referencing with genomic insights, and empathetic care agents like Sage closing the loop with patients at home. That’s the wave forming now, and it will feel less like discrete tools and more like a cohesive ecosystem.

Two years ago, nobody could have said for sure how far we would go in effectively integrating AI into healthcare operations and customer care. However, our predictions got pretty close. Here’s why: You can’t spend years leading innovation efforts for clients across countless industries without developing the toolkit of curiosity, informed dreaming, and practical systems knowledge that enables us to understand and wield AI. If you’re considering an AI pilot or are stuck in the early stages of AI strategy, we can help you develop and implement AI that actually works. Schedule a free discovery call, and who knows, maybe we’ll make some accurate predictions for you too.