In a guest post on Latent Space, Adrian Sanborn of Endura Therapeutics argues that AI has dramatically lowered the cost of thinking in science but has done little to speed up physical experiments. That asymmetry, he writes, is pushing biotech companies toward two different strategies: foundries and navigators.
Foundries, like Xaira, NewLimit, and Insitro, invest heavily in instruments and automation to generate experimental data an order of magnitude faster, with AI making that data legible and predictive. Navigators, by contrast, use AI within normal company operations to guide decisions and speed processes, without needing a proprietary model or massive dataset. Sanborn says navigation is less flashy but available to every company, and it runs fastest at early-stage startups with no legacy processes to slow them down.
Sanborn, who ran experiments at a bench during his PhD, also describes how faster code changes lab work itself. Adapting analysis to a modified protocol is now an afternoon's work rather than a project, so experiments can evolve more freely. Scientists who ran the experiments can present their own results instead of waiting on a computational specialist, removing a gatekeeper and opening up more hypotheses to exploration. He notes that lab software carries the opinions of outside vendors, and AI is shifting that balance so software can express a lab's own priorities.