Welcome to this week’s Research Agenda. Every Tuesday morning we’ll bring you the stories that matter most for UK research policy, as reported in full on the site, alongside everything worth a read elsewhere. This week’s essay looks at how artificial intelligence is changing how researchers collaborate, via a trip to San Francisco.

Plus: UKRI has set out its progress on addressing research financial sustainability while ARMA warns the sector is sleepwalking into losing research assets; Andy Burnham commits to saving Jodrell Bank, details tbc; Scotland gets a new innovation minister while ARIA loses its chair; we’ve charted ERC starting grants and the latest global talent visa figures; and Haldane muses about UKRI’s role in media controversies.

Michael Salmon, Editor

michael@researchagenda.news

Researcher in the loop

A new academic year, and everyone is talking about AI again. We’ve been here before. The capabilities of frontier models are at once revelatory and deeply alarming, and the summer press has been dominated by stories of cybersecurity breaches and rogue agent swarms. At the same time, the ability of artificial intelligence to push the boundaries of science and research marches ever onwards.

Earlier this summer we took a look at how AI is upending research in one particular subject area: mathematics. It’s surely a sign of the times that even in the few weeks since publication there have been further reported breakthroughs at the very cutting edge of maths research. One of the topics we’re interested to explore in Research Agenda is the specific disciplinary impact which AI is having on the research system, whether in use cases or community response. So if you have thoughts on the state of play in other fields, please drop us a line.

Beyond the individual subject area, there’s a wider question about how the plethora of new tools is reshaping researchers’ day-to-day work, careers, and professional identity, to say nothing of the well-rehearsed challenges around academic integrity and copyright. For some, the argument has long been that as long as academics and research staff are able to supervise the system – check the facts, assert authorship, take final responsibility – then the research system can proceed roughly as was. It’s becoming harder to sustain this line of thinking.

The idea of having a “human-in-the-loop” within AI systems – the dream of one central overseer who can check the quality and the ethics of what agents are up to – is becoming an unrealistic aspiration as complexity proliferates. And there’s also an increasing sense that this does not really represent a very fulfilling working life: rather than enjoying fully automated luxury research, the experience of being the operator is a rather fragmented and stressful one.

This week on Research Agenda, Holtzbrinck Group’s chief scientific officer Daniel Hook has been reflecting on what gets lost when AI smooths out the research process, arguing that there’s a real risk of “decollaboration” as the friction of dealing with other people becomes less central. He presents the case for retaining some of those human sticking points which make the process of thinking and writing what it is: the need to persuade colleagues, to reckon with genuine intellectual pushback, even to just slow down and wait for a response. Collaboration is not just a nice-to-have, he concludes – it’s a vital form of research infrastructure, and should be valued accordingly.

The week on the wire – every item reported in full on the site.

UCL Policy Lab director Marc Stears has been appointed as a special advisor to the Prime Minister, in a role aimed at “embedding the broader ecosystem of think tanks, academics and other experts into the business of government.”

A Conservative party reshuffle saw Claire Coutinho named shadow business secretary, with Julia Lopez as her maternity cover – and shadow science minister.

The Academy of Social Sciences has elected 78 new fellows, drawn from 38 UK universities, research institutes and other organisations as well as overseas institutions.

The British Academy/Leverhulme small research grants funding call is open, with up to £10,000 per award. A new resubmissions policy means that unsuccessful applicants will need to wait at least one full round before submitting the same proposal again.

The Association of Research Managers and Administrators has opened a survey on how research managers are using, not using, or considering using artificial intelligence in their work. The survey is open until 30 September.

The Innovation and Research Caucus has released a report on innovation funding, using Innovate UK data to interrogate what happens to firms that are successful in funding applications, as well as what happens with those that are unsuccessful.

Digital Science has opened its catalyst grant round for 2026, with a focus on novel applications of agentic AI across the research lifecycle.

Two UKRI research councils have published updated board meeting minutes: BBSRC and ESRC.

Worth your time from Wonkhe and across the wider community.

“Bizarre meaningless guff”: the triple helix explained  James Coe

What brass bands can tell us about REF reform  David Kernohan

We say we value diverse research, but can we assess it?  Ola Thomson and Gemma Derrick

Disruption eruption  Research on Research Institute

How one researcher captured a vanished world of UK government science  Nature

Universities rarely use research evidence in their decision-making, so why do they expect anyone else to?  LSE Impact

Where research policy made the news this week.

‘Exit tax’ on UK entrepreneurs ruled out by business department  Financial Times

Oxfordshire science facilities protected from funding cuts  BBC News

British fund management giants vie to run £1bn Scale-Up Fund  Sky News

A troubling rogue AI incident shows why the U.K. AI Security Institute deserves greater scrutiny  Fortune

The exact science that could be lost at Jodrell Bank, by a professor of cosmology in the know  Manchester Evening News