Intel / article

Foundries vs Navigators: Lowering the Cost of Science

Source: latent.space
Author: Adrian Sanborn
Date: Published Sep 24, 2026

Context

Endura Therapeutics' account illustrates how AI can widen the research considered before a drug company commits to experiments. Its chief executive describes a two-stage process: agents first prepared short assessments of about 500 disease targets, then produced deeper reviews of roughly 100 that remained. The second stage explicitly sought failed programs and reasons the company's approach might fare differently.

The reported benefit is broader early screening of possible research directions. The company says it checks deeper reports against primary sources and retains full human diligence for selected programs. Those checks matter because the account acknowledges errors in the generated reports; the process helps organize choices about experiments rather than establishing that a proposed treatment works.

Key takeaways · AI-distilled

Adrian Sanborn of Endura Therapeutics splits AI-era biotech into foundries, which cut the cost of experiments with high-throughput sequencing, microscopy, or automation, and navigators, which spend cheap AI thinking on what to run.
Because research protocols keep changing, fast AI coding lets the scientist who ran an experiment adapt its analysis in an afternoon, removing the analyst handoff that used to discourage protocol changes.
He argues labs should build their own data portals rather than buy vendor systems, since a tool's layout decides which comparisons are easy; implementation takes a day, deciding what it should show can take weeks.
Endura pointed research agents at about 500 disease targets: short reports filtered the first pass, then about 100 got 30-page-equivalent reviews prompted as a skeptical expert naming prior failures.
Sanborn notes limits: some reports had errors, tolerable because a first-pass miss only costs an opportunity, and analysis generated on demand has weaker than a versioned pipeline.

Terms in this piece · Glossary

LLM — A large language model — the neural network behind tools like Claude and ChatGPT, trained on huge amounts of text to predict what comes next.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
determinism — Whether the same input reliably produces the same output — something LLM systems mostly lack, which changes how you test and debug them.

Read the source www.latent.space