← All Journal posts
AI Research

The AI suggests. The scientist tests.

Jon Marrs, Ph.D. / ResearcherFlow//4 min read

On November 6, Cancer Research UK announced the next stage of its £8m AURORA project. A team at Harvard, led by Marinka Zitnik at Harvard Medical School, has built an AI "co-scientist" platform. Now three Cancer Grand Challenges teams will use it in their ongoing work and evaluate it as they go. NexTGen is working on CAR T-cell therapies for children’s solid tumors. PROMINENT is asking why some cells with cancer-linked genetic changes stay healthy while others turn cancerous. MATCHMAKERS is studying how T cells recognize cancer.

It’s a serious test, in real programs, on hard questions. But the sentence that stood out to me is about a limit.

"Can’t make any scientific decisions on their own"

The article describes the co-scientists as virtual lab assistants. They comb through papers and databases, spot patterns and gaps, and suggest the experiments most likely to teach you something. They’re designed to stick to reviewed evidence, and when there isn’t enough evidence to support a conclusion, they say so.

What they don’t do is decide. In the article’s words, "human scientists will still need to test it and evaluate the results before drawing any conclusions." Michelle Mitchell, Cancer Research UK’s chief executive, put it plainly: "it takes skilled scientists to turn a promising idea into a discovery that can make a difference to people’s lives."

That isn’t a hedge. It’s the design.

Legwork and judgment are different jobs

Most of a research week is legwork. Sweeping the literature. Turning papers into notes someone can find later. Reconstructing what happened in last Tuesday’s run. Writing the Monday update for the group.

That work matters, and it’s exactly the kind AI is getting good at. Judgment is a different job. Which hypothesis is worth a month of bench time? Does this pattern hold up, or is it an artifact of how the data was collected? What does a surprising result actually mean? Those calls belong to the people who know the field and will stand behind the answer.

The failure mode isn’t an AI that suggests bad ideas. Scientists suggest bad ideas too, and testing them is how understanding gets built. The failure mode is a system that quietly makes the call for you, or a week so full of legwork that nobody has time to make the call well.

How ResearcherFlow splits the work

ResearcherFlow handles the tedious parts of research, like literature sweeps, run logs, and the Monday writeup, so you spend your time on the science. It reads the papers, keeps the record, and proposes a next experiment with the evidence beside it. Nothing moves until you approve it.

Same principle the article describes, at the scale of a single lab. The system does the legwork and shows its sources. You decide what gets tested.

What we’re not claiming

ResearcherFlow has no connection to AURORA, Harvard, Cancer Research UK, or Cancer Grand Challenges, and we’re not comparing our product to theirs. The Cancer Grand Challenges teams are about to find out, in public, how well AI co-scientists work on real cancer questions. We’ll be reading along.

What we can say is that the principle (AI suggests, scientists test and decide) is how we think AI should work in every lab.

Literature in. Next experiment out. You approve.

Free to start with three workspaces, no card: https://researcherflow.com

Jon Marrs, Ph.D.
Founder, ResearcherFlow

Source: Tim Gunn, "Can AI 'co-scientists' help answer cancer’s biggest questions?", Cancer Research UK, November 6, 2026. https://news.cancerresearchuk.org/2026/11/06/can-ai-co-scientists-help-answer-cancers-biggest-questions/