The pipeline is filling up. Several drug candidates designed by artificial intelligence are now entering mid-to-late-stage clinical trials, and biotechs that bet early on the technology are beginning to look prescient. Iambic and Generate Biomedicines each expect to have three or more AI-designed drugs in human trials by the end of this year. For a field that has spent the better part of a decade promising to revolutionise medicine, the revolution is starting to look rather real.

This matters because drug discovery is, historically, a brutal business. It takes roughly twelve years and more than $2bn to bring a single drug to market, and nine out of ten candidates fail in trials. AI does not eliminate those odds, but it changes them. By modelling molecular interactions, screening millions of candidates and predicting toxicity before a single compound is synthesised, the technology compresses years of laboratory work into weeks. Nature recently highlighted one such method as capable of "turbocharging the hunt for new medicines." That is, for once, not marketing copy.

From screen to clinic

One of the clearest illustrations of what this looks like in practice comes from pancreatic cancer — one of the deadliest and most treatment-resistant malignancies known. Researchers used AI to design a novel molecule that significantly boosts the effectiveness of standard chemotherapy in treating the disease. Pancreatic cancer has a five-year survival rate below 15%. Any meaningful improvement is not an incremental win; it is a lifeline.

The approach is emblematic of a broader shift. Rather than screening libraries of existing compounds and hoping something sticks, researchers are now designing molecules from scratch — engineering them to hit specific targets with a precision that traditional methods cannot match. The chemistry is still hard. Getting from a promising molecule to a safe, manufacturable drug remains an immense challenge. But AI narrows the search space in ways that were simply not possible before.

Beyond the laboratory

The impact is not confined to drug design. AI is changing how diseases are detected in the first place, which determines whether treatments reach patients in time to help them.

Coronary microvascular dysfunction, or CMVD, is a heart condition that affects millions yet is notoriously difficult to diagnose. Current methods require expensive imaging. Researchers at the University of Michigan have developed an AI model that identifies CMVD from a standard ten-second electrocardiogram — the cheap, ubiquitous test that exists in virtually every clinic on the planet. Accuracy is high. The implications are significant: a condition that often goes undetected until it becomes dangerous could now be caught early, routinely, at minimal cost.

Separately, Chinese researchers have developed an AI framework that rapidly synthesises clinical data for evidence-based medicine. Systematic reviews — the gold standard for evaluating what treatments actually work — currently take months or years to produce. This framework compresses that process dramatically. Better, faster evidence means better, faster decisions about which treatments reach patients.

Hype, meet reality

None of this means the field has solved all its problems. It has not. Many AI-generated drug candidates will fail in trials, for the same reasons any drug candidate fails: unexpected toxicity, poor efficacy in humans, or a target that turns out to be less important than it seemed in the model. AI accelerates discovery; it does not guarantee success.

The Gartner Hype Cycle — a rough map of how technologies move from breathless anticipation to useful deployment — places medical AI at an instructive moment. After years at the "Peak of Inflated Expectations," the field is descending into the early "Slope of Enlightenment." That is actually good news. Technologies that survive that descent tend to become genuinely useful. The ones that do not simply disappear. Medical AI is not disappearing.

What the current moment represents is a calibration. The promises made five years ago were too large and too vague. The results arriving now are more specific, more modest and more credible. A molecule that improves chemotherapy outcomes. A ten-second test that catches a hidden heart condition. A faster way to know which treatments work. These are not the singularity. They are medicine getting incrementally, measurably better — which is exactly what medicine is supposed to do.

The year that counts

Whether 2026 becomes a genuine landmark depends on what the trials show. If several AI-designed drugs demonstrate efficacy in mid-to-late-stage studies, the case for the technology becomes difficult to argue with. Regulators will take note. Pharmaceutical companies that have been cautious will accelerate. Capital will follow.

If the trials disappoint, the correction will be sharp. The field has accumulated expectations it cannot afford to squander.

But the trajectory, for now, points upward. The molecules are in patients. The diagnostics are in clinics. The evidence is being generated. For a technology that has spent years being described as transformative, it is finally doing the work to prove it.