AI can design a better molecule, yet the clock on proving it works in humans barely moves [S1]. Dario Amodei reportedly speculated, in conversation with Dwarkesh Patel, that better AI drug design would let clinical trials take about one year [S1]. That prediction conflates two separate variables, molecule success rate and trial speed. Trial duration is set by enrollment, logistics, endpoints, and regulatory review (the FDA's structured process for authorizing a drug), none of which compress just because the drug candidate entering the pipeline is stronger [S1].
About nine of every ten drug candidates that reach clinical testing fail before approval, so design improvements start from a very low baseline [S1]. Most of that failure can be explained by 1) a lack of efficacy (40% to 50% of failures), 2) toxicity (30%), and 3) poor drug properties, the thing AI design most directly targets (10% to 15%) [S2].
Osteoporosis shows what happens even when the biology lines up. The field has an unusually strong preclinical model, the ovariectomized rat (a standard lab model for post-menopausal bone loss), which is preferred over other chronic-disease animal models but still has acknowledged limits in how closely it mirrors human disease [S3]. Despite that advantage, chronic-disease endpoints like fracture reduction (the measured outcome a trial must improve) require large, slow validation trials, so investment has slowed regardless of candidate quality [S1]. A good animal model does not buy you a short trial.
Zoom out and the pattern generalizes. The inflation-adjusted cost of bringing a new drug to market has roughly doubled every nine years, a trend known as Eroom's Law [S4]. The clinic is not only a filter that candidates pass or fail: trials generate the human data that improves the next generation of models, which then inform the next round of candidates [S5]. Starving that loop of trial-state investment starves future design too.
None of this means the operational bottleneck is fixed. Fortrea's regulatory-strategy work suggests early regulator interaction and adaptive pathways can reduce friction. AI design alone does not remove the bottleneck, though process design can mitigate it.
Investors and founders should focus on trial-state capability, patient recruitment infrastructure, endpoint strategy, and regulatory navigation as the means to expedite drug discovery, not just design velocity. The next funding cycle will separate teams that built for that reality from teams still selling one-year trials on the strength of a better molecule.
References
[S1] https://www.asimov.press/p/ai-clinical-trials
[S2] https://pmc.ncbi.nlm.nih.gov/articles/PMC9293739/
[S3] https://pmc.ncbi.nlm.nih.gov/articles/PMC2707131/
[S4] https://www.asimov.press/p/clinic-loop
[S5] https://www.asimov.press/p/clinic-loop