Anthropic says Claude can run science experiments now rather than just plan them

Boris Cherny, the creator of Claude Code, says he doesn’t write prompts for Claude anymore. “My job is to write loops,” the head of Claude Code at Anthropic has said, describing a shift away from manual, turn-by-turn prompting toward building small systems that keep an agent working until a defined condition is met. Peter Steinberger, the creator of OpenClaw and now a technical staff member at OpenAI, helped popularize the same shift: “you shouldn’t be prompting coding agents anymore,” he wrote. “You should be designing loops that prompt your agents.”

Anthropic is now predicting Claude will follow the same trajectory in life sciences. “Six months ago on this stage, we made a claim that Claude could help with the work of Life Sciences R&D,” said Zubair Jandali, the company’s head of healthcare and life sciences go-to-market, referencing Anthropic’s Briefing: Healthcare and Life Sciences livestream from January. “Today, we’re going to build on that claim with a new one: Claude can run the work, not help with it, not accelerate it, even run it. It’s a bold thing to say out loud, but let me tell you why we believe it. We’ve all seen this happen in software development. Coding has irreversibly changed.”

“The scientific method is a loop, too,” he said. “The original loop design: the experiment, run it, analyze the data, ask the next question. The experiment happens at the bench, the analysis happens at a keyboard, and that is where the loop stalls.”

Still, the word “loop” isn’t new to Anthropic’s life sciences vocabulary. At that January event, life sciences lead Eric Kauderer-Abrams used it too, telling the room Anthropic’s guidance was not to “let Claude run open-loop and write clinical trials and communicate directly with FDA and cut you out of the loop.”

But the loops are tightening. Jandali’s claim was the opening act for what Anthropic actually shipped that day at “The Briefing: AI for Science” in San Francisco. Claude Science runs on Anthropic’s existing Claude models, including Sonnet 5, introduced the same day. Claude Science bundles more than 60 scientific databases and connectors into one workbench. It includes a coordinating agent that spins up specialist sub-agents for individual tasks, and a reviewer agent that checks citations and calculations before handing results back. It runs on a lab’s own infrastructure, a laptop, a Linux box or an HPC login node, so large or sensitive datasets stay put, with only the context each analysis step needs sent to Claude. It’s in beta for Pro, Max, Team, and Enterprise subscribers.

Anthropic backed the push with more than a product launch. In April, it paid roughly $400 million in stock for Coefficient Bio, an eight-person startup founded by ex-Genentech computational biologists building AI tooling for drug-pipeline planning, target identification, molecular design, and clinical trial strategy. Alongside Claude Science, life sciences leaders Eric Kauderer-Abrams and Jonah Cool announced Anthropic would run its own internal drug discovery program, targeting neglected diseases that traditional biopharma companies wouldn’t pursue, framed as a way to build the company’s own development experience and credibility with the biopharma customers it’s selling Claude Science to. Kauderer-Abrams did not say what Anthropic would do with any promising candidates the program finds, the part of the process where a real biopharma company would move into clinical trials. Is Anthropic now a drug developer? Not really. It’s positioning itself to look like one to the customers it’s trying to sell to.

At the event, Anthropic CEO Dario Amodei joined Lotte Bjerre Knudsen, Novo Nordisk’s outgoing Chief Scientific Advisor and one of the scientists who spent more than three decades turning GLP-1 biology into Ozempic and Wegovy. STAT’s Matt Herper moderated.

Anthropic is now a drug hunter

In April, Anthropic paid roughly $400 million in stock for Coefficient Bio, an eight-person startup founded by ex-Genentech computational biologists building AI tooling for drug-pipeline planning, target identification, molecular design and clinical trial strategy. Where Claude Science wraps Anthropic’s existing Claude models in workflow tooling and database connectors, OpenAI built GPT-Rosalind as a model purpose-built for biological reasoning rather than a general-purpose assistant repurposed for science, with early users including Amgen, Moderna, the Allen Institute, and Thermo Fisher Scientific.

Alongside Claude Science, life sciences leaders Eric Kauderer-Abrams and Jonah Cool announced Anthropic would run its own internal drug discovery program, targeting neglected diseases that traditional biopharma companies wouldn’t pursue. The company framed the news as a way to build the company’s own development experience and credibility with the biopharma customers it’s selling Claude Science to. Kauderer-Abrams did not say what Anthropic would do with any promising candidates the program finds.