Hi!
Welcome to Signal, Valency's newsletter.
The AI-and-science conversation has been impossible to escape for the past year or so, from Big Tech incumbents to small academic groups, startups, and independent researchers. The models are impressive, but they were never designed with the rigor science requires: training data goes stale, citations get hallucinated, and papers in major publications have confidently cited papers that do not exist. That's why we started Valency.
Our first step is Valency Bond, an MCP service (the protocol that lets AI assistants call external tools) that plugs frontier or open AI models into an embeddings-indexed corpus of the scientific record: tens of millions of papers and preprints from 1965 through yesterday, updated daily. When you and your favorite model reason about science together, every claim should be grounded in something published, with the exact source.
Bond is still in beta. If you want to kick the tires, join the waitlist, and tell us what's hard about bringing AI into your day-to-day research. Grab a time with us.
Recent updates
[Blog] Navier-Stokes and the future of AI acceleration in science
Josh, our CEO, wrote about the news of a partial solution to the Navier-Stokes equations, one of the Millennium Prize problems, with the proof formalized in Lean. He argues it's a milestone for frontier models in the "thinking" layer of science, and that axiomatic reasoning is one component of a stack that runs up through compute, data, and new instruments.
[Blog] We will need more scientists, not fewer
Also from Josh, a piece on the question keeping a lot of researchers (and grad students especially) up at night: what happens to scientists when AI co-scientists get good at exactly the work we trained to do?
As AI pushes the frontier of knowledge outward, the volume of unexplored space grows faster than anyone can keep up. Choosing the right questions, staying skeptical of results that look too good, and owning the conclusions: that work stays with us.
[Blog] Science Won't Wait for Its Institutions
I wrote up notes from one of the salon dinners we've been hosting with researchers and leaders working across AI and science. The conversation had moved past whether the models are capable and toward what happens when research output accelerates beyond the systems built to evaluate it.
Math may be approaching an era of proof abundance, while biology still depends on experiments and mechanistic understanding. Across fields, the harder constraint is increasingly institutional: the technology is moving faster than the organizations that fund, publish, and govern science.
[Announcement] We joined the DOE's Genesis Mission
Valency was recently selected for a Department of Energy award through the Genesis Mission, joining Berkeley Lab on a project called HERALD. The TL;DR: decades of American nuclear power research sit locked in archives of millions of reports, drawings, and logs, because each one needs expert review before it can be released, and there are very few people qualified to do that review. HERALD gives AI agents the first pass through the archive, handling routine material and flagging what needs human judgment, with the experts in the loop. It's the first real-world deployment of our infrastructure. Here's the full announcement.
[Announcement] We joined the Agentic AI Foundation
Ryan, our COO, wrote up why Valency joined the Linux Foundation's Agentic AI Foundation, the body that has stewarded the Model Context Protocol since Anthropic donated it in December 2025. MCP is how Bond talks to every model we support, so we want a seat at the table as the protocol matures. Read the full piece →
We're hiring
The team is growing fast, and we have lots of open roles. If helping people make sense of the scientific record and do better research sounds like your kind of thing, take a look. Or just reply to this email and say hello. We read every email.
Come say hi in person
First stop: AGNTCon + MCPCon Europe, September 17 and 18 in Amsterdam. Avery and Theo will be at our booth demoing Valency Bond. AI Engineer Paris follows the week after, September 23 and 24. If you'll be at either, come talk to us about how you're incorporating AI into your science, or bring your strongest opinions about what we should be doing next. If you want to plan ahead, hit reply and we'll set something up. We'd also love to chat about running a workshop or giving a talk at one of your favorite conferences.
That's it for issue one.
Thanks for reading,
Karthik
You just read issue #1 of Valency Signal. You can also browse the full archives of this newsletter.
