Anthropic Now Owns a Wet Lab
Anthropic is running an actual biology lab in the Bay Area. Not a compute cluster with a biology team attached, a wet lab, with benches and reagents, where physical experiments get run. Eric Kauderer-Abrams, who heads life sciences there, put it plainly: to do biology, the final test is still real lab work, and they are doing that today.
The focus is fundamental biology, not drug discovery, which is a more interesting choice than it sounds. Drug discovery is where the money is and where every AI-bio startup goes, because the value is legible and the partners are obvious. Fundamental biology is where you go if you think the bottleneck is knowing how things work, not screening compounds faster. It is the slower, less fundable answer, which is roughly what you would expect from a company that can fund it out of a model business.
The pieces have been assembling for a while. Anthropic bought Coefficient Bio, a stealth AI biotech, in April for four hundred million dollars. They launched a Life Sciences Verification Program to hand vetted researchers access to the strong models. They have published on protein design acceleration and biomolecular modeling. There is a joint drug discovery partnership with Novo Nordisk. Now there is a building where experiments actually happen.
The reason to care is the loop. Everything in AI for science until now has been open-loop: the model reads the literature, proposes something, and a human decides whether to spend three months and a graduate student finding out. The model never learns what happened. Close that loop and you get the thing that actually compounds, a system that forms a hypothesis, runs it, and updates. That is the same structure that made agents work in software, where the compiler and the test suite provide the ground truth for free. Biology has no compiler. A wet lab is the closest available substitute, and it costs a great deal more than a CI run.
The uncomfortable part, which TechCrunch noted and which is fair, is the timing. Dario Amodei says AI could cure most major diseases in five to ten years and wants real-world testing to get there. The same company's researchers spend their public hours warning about what sufficiently capable models might do, with bio capability specifically near the top of the list. Both positions can be sincere and they still sit strangely in one building.
← Back to all articles
The focus is fundamental biology, not drug discovery, which is a more interesting choice than it sounds. Drug discovery is where the money is and where every AI-bio startup goes, because the value is legible and the partners are obvious. Fundamental biology is where you go if you think the bottleneck is knowing how things work, not screening compounds faster. It is the slower, less fundable answer, which is roughly what you would expect from a company that can fund it out of a model business.
The pieces have been assembling for a while. Anthropic bought Coefficient Bio, a stealth AI biotech, in April for four hundred million dollars. They launched a Life Sciences Verification Program to hand vetted researchers access to the strong models. They have published on protein design acceleration and biomolecular modeling. There is a joint drug discovery partnership with Novo Nordisk. Now there is a building where experiments actually happen.
The reason to care is the loop. Everything in AI for science until now has been open-loop: the model reads the literature, proposes something, and a human decides whether to spend three months and a graduate student finding out. The model never learns what happened. Close that loop and you get the thing that actually compounds, a system that forms a hypothesis, runs it, and updates. That is the same structure that made agents work in software, where the compiler and the test suite provide the ground truth for free. Biology has no compiler. A wet lab is the closest available substitute, and it costs a great deal more than a CI run.
The uncomfortable part, which TechCrunch noted and which is fair, is the timing. Dario Amodei says AI could cure most major diseases in five to ten years and wants real-world testing to get there. The same company's researchers spend their public hours warning about what sufficiently capable models might do, with bio capability specifically near the top of the list. Both positions can be sincere and they still sit strangely in one building.
Comments