GXL

We’re excited to sponsor Nucleate’s JPM ’27 Reception! Looking forward to connecting with the incredible community of scientists, founders, and builders shaping the future of biotech and AI in San Francisco this January!

Had your morning coffee? Here’s a scientific question to go with it: Does coffee consumption lower the risk of type 2 diabetes? Test your own claim at https://lnkd.in/eVR_9rBj

Finding papers about a scientific claim is not the same as finding evidence for or against it. A paper may mention a claim as background, cite an earlier result, or examine a related mechanism without actually testing the claim. Today we’re announcing the Paperclip Evidence API to make the relationship between a claim and its sources explicit. You can submit a scientific claim, and Paperclip will return the passages that support or refute it, linked back to the underlying literature. The result is is a structured view of the evidence that can be inspected, challenged, and incorporated into a larger research workflow at an incredible speed. In about one second, you can: - examine whether certain health claims are grounded in literature - get citations to back up claims - test autosearch hypotheses programmatically at scale Try it on a claim from your own work, and let us know what you think. https://lnkd.in/g7zQsEgn

Save the date: re:AGENT is back! GXL is hosting re:AGENT II, the hackathon for agentic science. The two-day build, from January 9th-10th, kicks off JPM week in San Francisco. Want to build or sponsor? Fill out the interest form in the comments Questions? Reach out to Vanessa Esquivel. Stay tuned for the official announcement with our collaborators, tracks, and more in the coming weeks!

2 days until the London AI x Science Hackathon! GXL is heading to London this weekend, October 3–4, to join some of the best researchers and engineers pushing frontier AI into hard science problems. Rafael Prévost-Rivero will be representing GXL on the ground. Catch him at Saturday's opening ceremony, where he'll show teams how to get started with Paperclip. He'll be around all weekend to help teams build.

In AI-driven drug discovery, the bottleneck is often access: agents can only reason over the scientific knowledge they can find and read. James Zou, co-founder of GXL, and his Stanford lab published a study in Science this month, highlighting the opportunity within existing scientific information. His team created a "virtual biotech" built from tens of thousands of AI scientist agents working across drug discovery and development. The architecture looks a lot like a drug developer: a chief scientific officer agent coordinating teams for target discovery, molecule design, safety, and clinical study design. One part of the system created 37,000 clinical-trial agents to analyze roughly 50,000 trials in under a week. A human team would need years. A few results worth noting: Drugs targeting genes that act like on/off switches in specific cell types were 40% more likely to advance from phase 1 to phase 2, 48% more likely to reach approval, and had 32% fewer adverse events. Using only data from before January 2025, the agents designed an antibody-drug conjugate against B7-H3 for lung cancer. Merck independently arrived at the same approach, and its drug later received FDA breakthrough therapy designation. The agents also improved when they had disagreement built in. Adding "devil's advocate" agents to challenge ideas led to better reasoning. So much of what the virtual biotech needed already existed, scattered across papers, trial records, regulatory filings, patents, and biological databases. Making that knowledge usable for agents at scale is becoming central to AI-driven R&D. That's the work we're doing at GXL. The agents in this study ran on Paperclip, our virtual filesystem covering 385M+ documents across literature, clinical trials, regulatory filings, patents, and biological databases. Agents can search across sources, read evidence in context, and trace findings back to the original record. Researchers at 12 of the top 20 pharma companies, 50+ biotechs, and 100+ universities use it today. If you're building with agents for drug discovery or repurposing, I'd love to hear what you're working on. Comment or send me a message. https://lnkd.in/gQqsBDYX

Congratulations to our co-founder and Chief Scientist James Zou and his Stanford team on being featured in The New York Times! Their new Science paper built a "virtual biotech": 37,000 AI agents that act like a real drug development company, from finding drug targets all the way to designing clinical trials. We are lucky to have him at GXL, where he's helping build the environments AI agents need to tackle the hardest scientific questions. Read the full story here: https://lnkd.in/gbwt3u6g #AI #DrugDiscovery #AIAgents #Biotech

GXL builds the environments AI agents need to tackle the hardest scientific questions, so we're proud to sponsor the MMBU Challenge. Spots are filling up quickly, so register here: https://luma.com/28k1tyd3

Congratulations to James and team! Tens of thousands of AI agents working like a real biotech, from target discovery to clinical trial design. The standout for us: 37,000 clinical-trial agents annotated 57,000 trials into an open, standardized database, showing that targets backed by single-cell evidence are 48% more likely to reach market. We're proud Paperclip could play a part, giving these agents fast, native access to clinical trial and regulatory data so they could reason over thousands of studies directly. Try Paperclip here: https://paperclip.gxl.ai/

GXL is officially part of the London AI × Science Hackathon, alongside Google DeepMind · Anthropic · Hugging Face · Originator · IQ Capital · Cambridge Compute Co. · Voyager Ventures · Modal · Amass · AIDDA. GXL builds the environments AI agents need to tackle the hardest scientific questions, so we are excited to be in this room. → Hosted by Iterate & future.bio → October 3–4th, London https://luma.com/3iipivod