FutureHouse previews Finch, an AI tool for biology discovery

Imagine handing a massive stack of cancer papers to a first-year graduate student, asking for molecular drivers of metastasis, and getting back executed code, freshly plotted figures, and parsed findings within five minutes.

That is essentially what Finch promises. FutureHouse, the Eric Schmidt-backed non-profit with a 10-year mission to build an autonomous AI scientist, just unveiled the new tool for data-driven biology discovery, hot on the heels of launching its platform and API last week.

Code, plots, and a few rookie mistakes

According to FutureHouse co-founder and CEO Sam Rodriques, Finch ingests biological data primarily from research papers, takes a research prompt, writes and runs code, generates charts, and inspects the output.

Rodriques called the ability to pull this off in minutes a superpower on X, likening the system to a talented first-year grad student that has already proven awesome on internal projects.

However, like any rookie researcher, Finch still makes silly mistakes. Because biology leaves virtually zero room for hallucinations, FutureHouse is keeping the tool in closed beta and actively recruiting bioinformaticians and computational biologists to stress-test its accuracy and train it further.

Big pharma bets meet harsh reality

FutureHouse is riding a massive wave of Silicon Valley optimism around automated science.

Earlier this year, OpenAI CEO Sam Altman claimed superintelligent AI would massively accelerate scientific discovery. Anthropic launched its own AI for science initiative this week, with its chief executive openly speculating that AI could help cure most cancers. The commercial stakes are enormous: Precedence Research estimates the AI drug discovery market will surge from $65.88 billion in 2024 to $160.31 billion by 2034.

Yet hard scientific evidence remains thin. Most working scientists still doubt AI can meaningfully steer the research steering wheel, and FutureHouse itself has yet to claim a novel scientific breakthrough using its own software.

The real world has been unforgiving. Drug discovery startups like Exscientia and BenevolentAI suffered high-profile clinical trial failures in recent years, while even flagship models like Google DeepMind's AlphaFold 3 continue to see wide swings in real-world accuracy.

Why it matters

For wet-lab researchers and computational biologists, Finch looks like a legitimate time-saver for grunt work. Automating literature synthesis, ad-hoc python scripting, and charting into a single prompt could shave days off early exploration, provided you treat every output with heavy skepticism.

Across the industry, this signals a shift from broad chatbots to domain-native autonomous agents that run executable code against specialized corpora. FutureHouse is smart to frame Finch as an error-prone junior researcher rather than an all-knowing oracle, setting grounded expectations where peers overpromise.

Still, we think running code and summarizing published papers is the easy half of biology. The real test is generating valid, testable hypotheses that human researchers missed. Until Finch proves it can find something truly new without breaking clinical reality, it remains a slick workflow accelerator rather than a genuine scientist.