It’s one of the most exciting and important times in pharma. It’s no secret that the integration of AI across industries is valuable, offering not only significant economical benefit, but changing the way humans operate. We’ve seen the implications in software, and as human intuition becomes stronger and models become better, I wonder about the one industry where, for the first time, AI feels inherently good.
“AI” has become a buzzword in today’s economy, with its meaning carrying a slightly negative connotation despite the operational efficiency and cost savings it can unlock. In healthcare, the impact is truly transformative. A few below.
1) One of the biggest bottlenecks we see is along the patient journey from prescription to fill, where the odds of losing a patient feel more like a coin flip than an edge case. Over 20% of patients are primary non-adherent: they never fill the prescription at all, often because of cost or insurance friction (PA delays and denied claims). Platforms like OpenEvidence, Hippocratic AI, and Tandem AI operate at this inflection point, offering healthcare agents to support both provider and patient along the process, automating benefits appeals, or creating an open source database with targeted medical literature. In today’s challenging healthcare economy, AI offers something rarer than just innovation. It offers access.
2) The second, and undoubtedly the most transformative, application is in drug discovery. For decades, human progress has been limited by one thing: knowledge. And like all good thought experiments, developing knowledge takes time. Go-to-market for an oncology drug can take up to 20 years once regulatory considerations are factored in, and even then, we won’t know if a patient responds or progresses on the therapy until one to two years in. Even then, we’ll have to wait decades to truly understand the long-term effects of these drugs on our biology, because DNA itself is a naturally evolving process.
That’s exactly what makes the tools diagnostics companies are building (tools that predict a patient’s prognostic and therapeutic response to a drug and improve patient stratification) so critical. Startups are creating novel platforms to study developmental biology ex vivo, so we no longer have to wait years to observe a natural biological response. We’re seeing this play out commercially, too: pharma giants are racing to own the hottest diagnostic technology, Roche acquiring PathAI, Tempus AI acquiring Personalis. Today, artificial intelligence is shaping the way humans operate, think, and now, evolve.
Training these models and ensuring accurate, reliable data has never carried higher stakes. In most industries, an inaccurate output means a lost quarter, a bad forecast, a few million dollars written off. In healthcare, it means a false positive that sends a healthy patient down a path of unnecessary treatment, or worse, a false negative that tells a sick patient they’re well. It is crucial, therefore, that the data underlying these models is validated and reinforced with the same rigor we’d demand of the clinicians and diagnosticians these tools are meant to support. Data is no longer just an input; it is the foundation of clinical judgment itself.
I urge researchers to continue pushing this frontier, and I urge physicians, patients, and policymakers alike to meet that innovation with equal urgency, welcoming not just AI, but AI safety, into the practice of medicine. Because if we get this right, the bounds of what’s possible in healthcare aren’t just wide. They’re truly unlimited.