Isomorphic enters the clinic — and why data, not models, will be the moat
Isomorphic Labs gears up for human trials, Owkin's "AI scientist" lands at three top-ten pharmas, and Bessemer makes the case that biology-native data is the only durable advantage.
The week the AI-drug-discovery story stopped being about benchmarks. Isomorphic Labs is publicly preparing its first human trials, Owkin's autonomous "AI scientist" has shown up inside Sanofi, BMS and Merck, and a sharp Bessemer essay argues — convincingly — that biology-native data, not models, is the only moat that holds.
Lead — Isomorphic's clinical inflection point
Isomorphic Labs is preparing its first human trials with AI-designed molecules. The interesting signal isn't the headline — it's the timeline slippage and what it reveals about the clinical-to-model feedback loop pharma still hasn't built well. If your AlphaFold spinout doesn't see assay-level data quickly, the model degrades into something prettier than it is useful.
Owkin's "AI scientist" goes agentic at scale
Owkin's autonomous research agent is now in production at Sanofi, BMS and Merck. The framing matters: it isn't a copilot, it's a teammate that proposes hypotheses, queues experiments and reads results. Expect the org-design conversation inside R&D to get awkward — fast.
Bessemer: data infrastructure is the only moat
Bessemer's new thesis is that "biology-native" data infrastructure — owned, standardised, structured for retraining — will be the defensible advantage in pharma AI, not the model layer. I agree, with one caveat: the moat is data plus the assays that generate it. Without lab control, you're still renting a moat.
Quick hits
Alloy Therapeutics raises a Series E pegged to AI-augmented antibody platforms. New Phase-1 success-rate claims out of two AI-first biotechs (read the methodology, not the headline). NVIDIA expands BioNeMo to cover small-molecule property prediction. EU AI Act compliance deadline tightens — pharma legal teams should be re-reading Annex III.