AI is helping Endura Therapeutics analyse experiments, build research tools and screen hundreds of disease targets, according to Adrian Sanborn in Latent Space. The important shift is not a miraculous robot scientist, but a practical one: researchers can now explore far more options before committing scarce laboratory time.
Watch Desk analysis
What happened
Sanborn, who has a computer-science background and experience running experiments at the bench, describes two ways AI is changing scientific work. “Foundries” make experiments faster through industrial-scale measurement and automation. “Navigators” use AI inside ordinary research operations to decide what to analyse, which tools to build and which questions deserve an experiment.
At Endura, language-model agents screened roughly 500 disease targets in a first pass, then examined about 100 of the survivors in greater depth. The system produced research reports covering disease biology, existing treatments and previous programme failures. Human researchers still check the results against primary sources, and selected programmes receive the full diligence process.
The article also describes internal software built around the company’s own experiments. Scientists can adapt analysis pipelines as protocols change, create data dashboards in hours rather than days, and inspect results without waiting for a separate computational team to interpret every measurement.
Why it matters
Scientific research is not held back only by a shortage of ideas. Physical experiments can take days or weeks, and a promising theory still has to survive contact with cells, instruments and reality’s famously uncooperative furniture. AI can make reading, coding, comparison and triage dramatically cheaper, but it does not remove the need for reliable measurements.
That changes where advantage may sit. A company does not necessarily need a uniquely trained model if it can use general models to ask better questions, adapt its tools quickly and reject weak research directions earlier. Sanborn argues that this “navigation” is spreading quietly because it changes how teams work rather than producing a glossy product announcement.
Our read
The most convincing claim here is also the least cinematic. AI may matter in science because it helps researchers decide what not to do. Screening 500 possibilities does not create 500 medicines, and model-generated expertise is not a substitute for experimental evidence. But making broad, informed triage possible could reduce the number of expensive mistakes that survive simply because nobody had time to look closely.
Readers should watch the hand-off between model output and laboratory proof. The interesting question is not whether an agent can write a long report. It is whether its recommendations improve the quality, speed and success rate of the experiments that follow.
What to watch
- Whether Endura’s AI-assisted disease triage produces drug candidates that survive laboratory and clinical testing.
- How often human researchers overturn the agents’ conclusions after checking primary sources.
- Whether other biotech firms build comparable internal tools rather than buying generic platforms.
- Whether faster analysis leads to better experiments, or merely more hypotheses waiting in the same physical queue.
Discussion spark: If AI makes it cheap to generate and rank hundreds of scientific hypotheses, should research funding shift towards automated experiments and verification rather than more idea generation?
Sources and evidence
- Foundries vs Navigators: Lowering the Cost of Science (24 September 2026, 15:03 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.