Lila's Autonomous Science Factories
Lila Sciences
The key divide in AI drug discovery is no longer who has the best model, it is who can turn model output into fresh experimental data fastest. Isomorphic can propose targets and molecules using AlphaFold and Alphabet compute, but when validation sits with outside labs, each design cycle depends on partner queues, sample handoffs, and external assay capacity. Lila is built around owning that loop inside autonomous facilities, which makes speed and data capture part of the product.
-
In practice, external validation means the AI company designs an experiment, then a CRO or pharma partner runs the assay and sends results back. That keeps fixed costs lower, but it also slows iteration and limits control over exactly how data is generated, labeled, and fed back into the model.
-
Lila is taking the opposite approach. Its AI Science Factories combine models, robots, instruments, and data systems in one closed loop, so the same system that suggests an antibody or catalyst can also execute the protocol, measure the outcome, and launch the next round of experiments.
-
The closest comparables show the tradeoff. Recursion and Insilico have both invested in automation and internal data generation, because owning wet lab throughput creates proprietary datasets and lets them capture more value than a software only discovery vendor. Benchling sits one layer over this stack, as the system of record for lab work rather than the lab itself.
The market is moving toward tighter integration of computation and experimentation. Companies that own more of the physical validation loop will train on better feedback, shorten discovery timelines, and look more like industrial R&D platforms than software vendors, which is why autonomous facilities are likely to become a central competitive moat in AI driven discovery.