Life sciences companies sit on three operating models in one: an R&D engine, a manufacturing operation, and a commercial business. Intelligent automation interacts with each differently. Treating them the same is the most common mistake.
R&D: augmentation, not autonomy
In R&D, the highest-value work is scientific judgement. AI as a co-pilot accelerates literature review, hypothesis generation, and experimental design. AI as an autonomous decider does not work and should not be tried. The AI Factory operating model for R&D keeps the scientist at the centre and uses agents to compress the cycle time of every iteration.
Manufacturing: rule-driven and high-stakes
Manufacturing is rule-driven, audit-grade, and intolerant of variability. Automation here is well-trodden: MES, batch records, deviation management. Agentic AI adds value at the analytical layer, identifying patterns in deviation data that human operators do not have time to see. Senior operations leads hold the GxP gate.
"The single highest-leverage change in life sciences is to build the operating model around the agent, not to bolt the agent onto the operating model."
Commercial: where AI moves the P&L fastest
Commercial functions move the P&L fastest because the regulatory gate is lighter. Market access, field force productivity, customer engagement. The economics here look like consumer AI more than pharma AI: agentic personalisation, decision support, content generation at scale.
The takeaway. Life sciences is not one AI problem. It is three. The companies that segment the operating model question by function, and rebuild each one separately around the right level of AI autonomy, are the ones compounding the advantage. Single-model approaches stall at pilot.