AI in healthcare has been “about to transform everything” for roughly a decade now. Every year brings new headlines about breakthrough algorithms, every conference features panels on the AI revolution, and every health system has at least one AI initiative underway. But what’s actually working in clinical practice today? And what’s still more promise than reality?

As someone who works in sleep medicine — a field directly affected by AI developments — I’ve tried to keep an honest eye on the broader healthcare AI landscape. Here’s where I think things genuinely stand as of early 2026.

What’s Working Well

Radiology AI has arrived. Not in the “replace radiologists” way that was predicted a few years ago, but in a meaningful clinical support capacity. AI tools for detecting pulmonary nodules, mammographic abnormalities, intracranial hemorrhage, and fractures are deployed in hundreds of hospitals worldwide. They function as a second reader — flagging findings that might be missed and prioritizing urgent cases in the reading queue.

The FDA has cleared over 800 AI-enabled medical devices, with radiology representing the largest category. Real-world performance data is accumulating, and the results are generally positive: AI catches things humans miss, and humans catch things AI misses. Together, they’re better than either alone.

Pathology is following a similar trajectory. AI analysis of tissue slides is gaining traction, particularly for identifying cancer subtypes and grading tumor aggressiveness. The workflow integration challenges are being solved, and some large pathology groups now use AI routinely.

Clinical deterioration prediction is another success story. Systems like Epic’s Sepsis Model (despite early controversy) and various early warning scores have been refined and deployed across health systems. When properly calibrated and integrated into nursing workflows, they demonstrably reduce response times to deteriorating patients.

Administrative AI might be the least glamorous but most impactful application. Ambient clinical documentation — AI that listens to the doctor-patient conversation and generates clinical notes — is reducing documentation burden for physicians. This matters enormously for burnout, which remains a crisis across medicine.

What’s Overhyped

AI-driven drug discovery generates fantastic headlines but hasn’t yet produced the revolution that was predicted. Several AI-designed molecules have entered clinical trials, which is genuinely noteworthy. But the timelines from molecule to approved drug remain measured in years, and the success rates in clinical trials haven’t dramatically changed. AI is accelerating early discovery, but it hasn’t fundamentally altered the pipeline.

Chatbot-based patient triage was supposed to reduce emergency department volumes and primary care wait times. In practice, most health system chatbots still frustrate patients with generic advice and excessive liability-driven caution. The technology is improving — large language models have made conversations more natural — but trust and accuracy haven’t caught up with expectations.

Precision medicine AI remains more promise than practice for most conditions. Outside of oncology, where genomic profiling genuinely guides treatment selection, the idea that AI will create perfectly personalized treatment plans based on your individual data is still largely aspirational.

Where Sleep Medicine Fits

Sleep medicine is an interesting case study for healthcare AI because the field has several characteristics that favor AI adoption.

First, our primary diagnostic test — polysomnography — generates highly structured data that algorithms can process effectively. AI scoring of sleep studies has reached accuracy levels comparable to experienced sleep technologists, and several products are commercially available.

Second, our major treatment — CPAP — generates continuous adherence data through cloud-connected devices. This creates opportunities for AI-driven monitoring and early intervention with non-adherent patients.

Third, the supply-demand gap in sleep medicine is severe. There aren’t enough sleep specialists to see every patient who needs one. AI that helps triage referrals, automate scoring, and streamline workflows can meaningfully expand access.

I’m cautiously optimistic about AI’s role in sleep medicine specifically, even while I’m more measured about some of the broader claims.

The Implementation Gap

The biggest challenge in healthcare AI isn’t building good algorithms. It’s implementing them in real clinical environments. This implementation gap is where most promising AI tools go to die.

Successful implementation requires:

• Integration with existing EHR systems (which is technically painful)
• Clinician buy-in (which requires education and demonstrated value)
• Workflow redesign (which requires operational expertise)
• Ongoing monitoring and maintenance (which requires sustained investment)
• Clear governance and accountability structures

Organizations like team400.ai are helping healthcare providers navigate this implementation challenge, recognizing that the last mile of AI deployment is often the hardest and most important.

Honest Concerns

I’d be doing a disservice if I didn’t mention the legitimate concerns about healthcare AI.

Bias remains a real problem. AI models trained on data from predominantly white, affluent patient populations may perform poorly for other groups. The dermatology AI that’s excellent at diagnosing skin conditions on light skin but misses them on dark skin is a real example, not a theoretical worry.

Transparency is often lacking. Many AI systems are “black boxes” — they produce outputs without explaining their reasoning. For a physician who needs to understand why a particular recommendation was made, this is a significant barrier to trust and clinical utility.

Data privacy concerns are growing as health systems share more data with AI vendors. Patients have legitimate questions about who’s training models on their medical information and for what purposes.

Workforce effects are uncertain. While the current consensus is that AI will augment rather than replace healthcare workers, the long-term impact on training pipelines, skill development, and employment is genuinely unknown.

What Comes Next

My prediction for the next 2-3 years: we’ll see less hype and more quiet, practical deployment. The flashy “AI will replace your doctor” stories will fade, replaced by pragmatic applications that make healthcare incrementally better — shorter wait times, fewer missed diagnoses, less paperwork, more efficient operations.

The transformation won’t feel like a revolution. It’ll feel like your hospital gradually getting a little better at everything, year after year. And honestly? That might be exactly what healthcare needs. Not a disruption, but a steady improvement powered by technology that works quietly in the background while clinicians focus on what they do best — caring for patients.

The organizations that succeed with healthcare AI will be the ones that treat it as a tool in service of clinical care, not as an end in itself. That’s a boring conclusion, but it’s the right one.