Why AI in Life Sciences Is Still a People Problem
The models are good enough. The bottleneck is trust, workflow integration, and knowing what questions to even ask.
Everyone in pharma and biotech is running AI pilots. Most of them will not survive contact with the organisation.
Not because the technology fails — it usually works. The model predicts, classifies, generates. The problem is what happens next. A scientist looks at the output and asks: should I trust this? And without a good answer to that question, the pilot stalls, the champion moves on, and the slide deck gets archived.
This is the real frontier in AI and life sciences. Not model accuracy. Not compute. Not even data quality, though that matters. It is the human layer — the expertise needed to ask the right questions, interpret outputs with appropriate scepticism, and redesign workflows around what AI is actually good at.
What I keep seeing
The organisations making real progress share a few traits. They have scientists who are genuinely curious about AI, not just compliant with it. They treat AI outputs as a first draft, not a verdict. And they have someone in the room who can speak both languages — who understands the biology and can interrogate a model’s reasoning.
That last part is rarer than it should be.
The opportunity
If you can be that person — or build a team of those people — you are not competing on AI capability. You are competing on the ability to turn capability into outcomes. That is a much more durable advantage.
The models will get better and more accessible. The judgment layer will not commoditise at the same rate.