Everyone Will Be a Builder: What Separates AI Leaders From the 9% in Pharma

Only 9% of life sciences companies see measurable business impact from AI, and only 22% have scaled it at all, according to Deloitte's 2026 Life Sciences Outlook. Boris Bogdan presented those numbers yesterday at the AWS Life Sciences Symposium in Zurich, and his diagnosis of the gap was the best summary of where we stand that I have heard this year.

MIT scores healthcare and pharma at 0.5 on real AI disruption, four times lower than media. Meanwhile Insilico Medicine eyes a Q4 start for late-stage trials of an AI-generated IPF candidate, and Generate:Biomedicines is initiating global Phase 3 studies for an AI-engineered asthma antibody. The frontier is sprinting while the industry average crawls.

His explanation matches what I see in my advisory work. Companies stuck at pilot stage keep funding sexy demos. Companies pulling away invest in the unglamorous layer underneath: enterprise ontologies, metadata and context, semantics, dedicated agent infrastructure. Without that foundation, no use case scales past the proof of concept.

Three points that stuck with me:

Run two tracks. Harvest fast ROI cases now, like report automation, MLR pre-screening, and competitive intelligence agents, while the foundational layer gets built in parallel. One buys patience for the other.

Our data is optimized for human consumption. Agents need data optimized for machine consumption, including the tacit knowledge sitting in experts' heads. That migration is the real work, and research will need an automated lab in the loop on top of it.

Eli Lilly showed the destination. Their Cortex platform runs 52 models and serves roughly 34,000 active users, including 1,550 internal developers, at about 1 million prompts per month.

I think my most important message coming out of this was: everyone will be a builder in the future. Bench scientists, medical writers, supply chain managers, all of them listed on screen.

People can only build on top of what exists. Give a medical writer clean metadata, shared semantics and a real agent infrastructure, and she can assemble her own MLR pre-screening agent in an afternoon. Without that layer, she is back to filing a ticket with IT.