Drug discovery is one of the great paradoxes of modern science: astonishingly sophisticated, brutally slow, and punishingly expensive. The biology is complicated. The chemistry is unforgiving. Even when everything appears right in the laboratory, most drug candidates still fail in clinical trials. That long chain of uncertainty shapes the entire industry—what gets funded, what gets ignored, and which diseases remain stubbornly untreated.
A few weeks ago at Techonomy 25, I sat down with Adam Lewis, the Head of Innovation at SandboxAQ, for a conversation I’d been looking forward to all season. Adam’s work begins with a simple premise: maybe we can understand the underlying systems well enough to stop making so many bad bets.
A Different Kind of AI
The first thing Adam wanted to clarify was what SandboxAQ is not doing. They are not building larger language models and expecting them to understand biochemistry suddenly. “There’s a world of thoughts and symbols,” he said, “and then there’s a world of stuff—the things you can trip over.” Drug discovery happens in that second world. Molecules obey physics, not probabilities.
SandboxAQ refers to its approach as Large Quantitative Models (LQMs). They’re not asked to produce text or images. Their job is to represent atoms, proteins, and tissues with enough numerical fidelity that you can begin to predict what will happen before you run the experiment. In other words, the model doesn’t guess—it simulates.
That distinction sounds academic, but in practice, it reframes the entire workflow. Today, biology and chemistry operate like separate disciplines: biologists choose a target, chemists design molecules to hit it, and years later, clinical teams determine whether the original hypothesis had any connection to reality. Adam is trying to compress that into a single, iterative loop.
“If I take a molecule and a cell and put them together in a lab, I’ve now used them up,” he said. “But if I have them represented as information, I can do it all at once.” Computation doesn’t replace the experiment; it multiplies it, allowing biology to speak to chemistry and vice versa in a feedback cycle that the physical world can’t match.
Source: BIO, Informa, QLS Clinical Development Success Rates
The Real Bottleneck
For decades, the industry has been convinced that the significant inefficiencies are in the chemistry—the hunt for a molecule with the right properties. Adam doesn’t buy that. We’re actually decent at making molecules do what we ask. The deeper issue is that we often ask the wrong thing.
Drug programs frequently fail not because the molecule is flawed, but because the underlying biological assumption was incomplete or incorrect. When a drug unravels late in development, it’s usually because the disease is more complex than anticipated, the mechanism doesn’t hold across real patient populations, or the intervention hits the wrong node in a larger network.
Adam sees an opportunity in those early, murky questions—the part of drug development where intuition, experience, and incomplete data often guide billion-dollar decisions. If simulation can illuminate which targets are plausible and how they’re interconnected, then everything downstream becomes more coherent. You avoid the dead ends before you invest a decade in them.
SandboxAQ isn’t trying to become a pharmaceutical company. Instead, it embeds its technology into existing pipelines through partnerships. Sanofi is using its models to identify biomarkers and better understand the mechanisms of drugs already in development. Bahrain’s Mumtalakat fund has formed a collaboration to accelerate treatments for conditions like diabetes and inherited disorders—problems that are regionally acute and economically urgent.
SandboxAQ isn’t just about speed; it’s about empowering the industry to ask better questions earlier, inspiring confidence in the approach, not merely the potential to improve drug discovery.
That subtle change has real implications. Many of the most devastating diseases—neurodegenerative disorders, heart disease, stubborn cancers—don’t hinge on a single target. They involve webs of interactions across pathways and tissues. Modeling those networks quantitatively, rather than one molecule at a time, could finally move us past the single-target mindset that has limited traditional cancer.
Across the field, the momentum is unmistakable. Dozens of companies are now positioning themselves as AI-native drug-discovery firms. Roughly 30 AI-generated or AI-guided molecules have entered clinical testing. Several early programs have cleared Phase I, and case studies suggest higher-than-expected success rates at that earliest safety stage. No AI-designed drug has yet earned full approval, but the trajectory has shifted: these systems are now touching real pipelines, not just academic benchmarks.
Regulators have also begun to acknowledge their role. In late 2025, U.S. and European agencies cleared an AI model for evaluating liver disease biopsies in clinical trials—one of the first examples of an AI tool becoming part of the formal evidentiary process in drug development. Detailing the validation and regulatory acceptance process can help the audience understand AI’s credibility and future role.
Adam is careful not to oversell. The tools aren’t magic, and they don’t eliminate the need for wet labs or human insight. Biology will always reveal surprises. Explaining current limitations and the importance of human expertise can help manage expectations and foster a balanced understanding of AI’s role.
Why This Matters
What struck me most in our conversation was the balance of optimism and restraint. He sees the humanitarian upside—better treatments for under-researched diseases—as clearly as the commercial one. If AI can reduce risk early in the pipeline, suddenly it becomes viable to pursue diseases that don’t promise blockbuster returns. Rare conditions, geographically concentrated illnesses, and complex multi-pathway disorders all move into the realm of possibility.
There are a few places where AI could have a more tangible impact on human well-being. Not by replacing scientists, but by giving them better starting points. Not by conjuring new drugs out of thin air, but by helping us understand the systems we’ve been trying to fix for a century.
Watch our full interview below.
