We are betting on two tailwinds.
- Hypotheses are becoming abundant. AI agents and design models now propose targets, generate molecules and plan experiments in hours, at almost no marginal cost. The number of ideas worth testing is growing faster than ever, and soon most of them won't come from people at all.
- Evidence remains scarce. The instruments, protocols and expertise to test these ideas already exist across CROs, core facilities and specialist labs. But reaching them still runs on cold emails, PDF quotes and procurement cycles built for big pharma. A model can design a hundred binders before lunch; getting ten of them tested takes months of back-and-forth.
Scientists without a wet lab of their own hit the same four walls:
- Speed. Weeks of scoping and quoting, one vendor at a time, before any work starts.
- Access. CROs prioritize large, repeat accounts, so small, one-off orders get slow replies or none.
- Ops. An NDA, MSA, vendor onboarding and purchase order for every new lab, with no procurement team to absorb them.
- Closing the loop. Results come back as PDFs in each lab's own format, cut off from the hypothesis that prompted them, so the model that designed the experiment can't learn from it.
We are starting with AI-designed molecules (binders, peptides and engineered proteins), inside the AI research agents where those hypotheses are made. They are the fastest-growing source of testable ideas, and the assays that validate them are standardized enough to run through a network and fast enough to close the loop in weeks. A scientist or agent describes what it wants to learn; Litmus picks the assays, finds a vetted lab, handles contracting and payment, and returns structured results ready for the next design cycle.
From there, Litmus becomes an API for the physical world: every assay callable and priced, with agents as first-class customers. An agent that finishes a design run overnight can have samples on their way to a lab by morning, and one scientist can run a dozen design-test loops at once. Every experiment adds to a record of what was predicted and what actually happened, the ground truth the next generation of design models needs. AlphaFold needed the Protein Data Bank; the models after it need data that doesn't exist yet.
We are a small team, incubated by Fulcrum Science, betting that as intelligence becomes abundant, progress will be set by how quickly ideas meet reality. AI is making ideas abundant. Litmus makes evidence abundant too.