Drug discovery will not get cheaper until it leaves the bench. I think that is the right bet. Max Jaderberg told No Priors: AI, Machine Learning, Tech, & Startups that Isomorphic Labs is trying to move design work onto a computer with what he called experimental accuracy.
The failure happens late.
Jaderberg said things fail in the clinic because "we never could see the problems until we get to the point where we put these drugs in patients" and pointed to toxicity teams did not predict. The host said approvals are roughly flat while cost grew from a few hundred million dollars to about $1.5 billion and timelines stretched from eight or nine years to 15 years.
Jaderberg traced conviction to AlphaFold plus game systems like AlphaGo and AlphaStar, work he did at DeepMind after selling his vision startup there in 2014 and helping found Isomorphic Labs in about 2021, he said. The team runs about five AI staff to one domain expert and holds models to a dotted line for experimental accuracy on internal leaderboards. Isomorphic Labs remains preclinical, with first trials targeted by the end of 2026, as R&D World reported.
He concedes the hard part. Pharma treats each program as a bespoke project, he said, so learnings do not compound and old data stays biased. That is fair. No simulation has yet called the toxicity that only shows in patients.
Still, the platform claim survives because scale brought generality before. If you are building in bio, copy the ratio and the leaderboard. Do not pay for wet lab turns you could have simulated.
