Skip to content

Build1 publisher2 min readPublished

Claude Science lead engineer Alec Tarashansky joins OpenEvidence to work on AI for medicine

Alec Tarashansky, lead engineer on Anthropic's Claude Science, has joined OpenEvidence, the physician search company valued at $12 billion in January. The part of Claude Science's design that fits most directly into a doctor's search tool is its citation-checking agent.

The Engineer · Build desk

Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction

Illustration accompanying Claude Science lead engineer Alec Tarashansky joins OpenEvidence to work on AI for medicine
Generated illustration

What happened

  • Tarashansky wrote on X that he hopes to bring the approach behind Claude Science to doctors at the point of care.
  • Anthropic launched Claude Science as a beta on June 30, connecting Claude to more than 60 scientific skills and data sources and to local or remote compute.
  • The workbench saves the code, environment and conversation history behind each figure, and an agent checks its citations and calculations.
  • OpenEvidence sells AI medical search and clinical decision support to verified physicians, with answers grounded in medical literature and guidance.

Compiled by The EngineerSomething wrong?How this is made

Why it matters

  • constraint Saved code and compute environments have little to attach to in a retrieval product, so most of Claude Science's provenance layer has no obvious use at OpenEvidence.
  • exposure A citation checker inside OpenEvidence would sit under answers physicians use in patient care, so its misses carry more risk than in a research literature review.
  • contradiction Anthropic's August seat program for 10,000 scientists and its September lab announcement cut against reading this departure as Anthropic stepping back from science.

Claude Science keeps a provenance record because the product runs code. Saving the code and environment behind a figure [7] makes sense when the app sends jobs to local or remote compute [4] for genomics or protein-structure analysis [5]. In those fields a result depends on the pipeline that produced it. I think keeping the pipeline next to the figure is the correct design, and it is good engineering. Anthropic's published examples include single-cell analysis and long-form literature reviews [15].

A physician search tool is built on retrieval. RuntimeWire describes OpenEvidence's job as helping doctors retrieve and assess information relevant to patient care [8]. That loop has little code or compute environment to preserve, unless OpenEvidence plans to hand physicians a notebook kernel. Grounding answers in literature and guidance [8] is citation work. Claude Science already runs an agent that checks citations and calculations [7].

Tarashansky's own framing is about the user. In his post on X, he wrote that Claude Science was built so that "scientists' curiosity, not their capacity to execute" sets the limit on discovery [2]. For that idea to carry over to medicine, doctors would have to be limited by the work of processing evidence, the way he says scientists were limited by the work of execution [2][3]. His post does not describe a product or a role, and OpenEvidence's funding news does not describe his remit [9][11].

The workbench was a team build. Anthropic named Tarashansky lead engineer at an August 20 product event, and its launch materials describe a larger system and team [6]. He joined Anthropic as a software engineer on its life sciences team after three and a half years at the Chan Zuckerberg Initiative [12]. Before that, Stanford had him down as a Bio-X fellow studying how stem-cell lineages compete inside tissue [13]. He announced the OpenEvidence move 95 days after the Claude Science beta launched [1].

The evidence that builders of traceable workflows are moving into clinical tools comes down to this one engineer. Anthropic is still adding to its science work. In August it announced discounted or free Claude seats for 10,000 scientists, and in September a life sciences research group and laboratory [14]. OpenEvidence's valuation came from a $250 million Series D on January 21, co-led by Thrive Capital and DST Global, according to TechCrunch's report as cited by RuntimeWire [10].

What to watch

  • Whether OpenEvidence ships a citation- or calculation-checking agent in its physician product, and whether Tarashansky's name is attached to it.
  • Who Anthropic names to lead Claude Science engineering, and whether the app leaves beta.
  • Further moves by Anthropic life-sciences builders into clinical companies, which would turn one hire into a pattern.
Loading claim ledger
Loading source directory links
Loading share composer
Loading topic controls
Loading related stories