AI tackles pharmaceutical R&D's decade-long cost spiral

The summary frames AI as a solution to Eroom's Law, but the harder problem is data quality and proprietary lock-in: pharma companies that generate the richest experimental feedback loops will have a structural moat that no general-purpose model can easily replicate, regardless of algorithmic advances.

The agent coordination piece covered the same week under 'The path to artificial superintelligence' is directly relevant here. Drug discovery pipelines are exactly the kind of multi-step, multi-specialist workflow where isolated AI tools hit the coordination ceiling MIT Technology Review described: a model that predicts binding affinity cannot hand off cleanly to one that models toxicity if the two systems cannot share intermediate reasoning. Closing the data loop, as this story frames it, is partly a coordination problem, not just a modeling one. That connection suggests the timeline for meaningful AI impact in pharma depends less on any single model's performance and more on whether the orchestration infrastructure matures in parallel.

Watch whether a major pharma firm (Pfizer, Roche, or a mid-size biotech with a disclosed AI partnership) reports a measurable reduction in Phase I candidate attrition rates within the next 18 months. That would be the first falsifiable signal that closed-loop AI is doing more than accelerating early screening.