Could agentic AI topple grant-funding systems?

Writing a Comment piece in Nature,  Geraint Rees (UCL) and James Wilsdon (RoRI/UCL) highlight how this new breed of AI tools could fundamentally upend how research is funded and provide recommendations as to how funders can adapt.

AI agents – also referred to as agentic AI – are more advanced capabilities within large language models that don’t respond to a single prompt but pursue goals across multiple steps. They can search the web, read documents, write and execute code, call external services and more to deliver a specified goal.

When it comes to writing a grant proposal, AI agents can be trained on a researcher’s publicly available body of work, grant criteria, and previous successfully funded grants to generate ideas and even fully formed applications. Because of this, a seemingly high-quality grant proposal can be created in a tiny fraction of the time it once took, with minimal effort.

This runs the risk of overwhelming funding agencies with huge volumes of high-quality submissions to assign to a limited number of awards, who will have to make largely arbitrary choices about what or who to fund.

Lead author Professor Geraint Rees, UCL Vice-Provost of Research, Innovation & Global Engagement said: “Funding panels have always faced hard choices, but they could at least claim to be distinguishing excellent ideas from merely good ones. Agentic AI is making that claim increasingly hollow. Funders aren’t facing a distant threat — the data suggest the system is already under strain. The good news is that better approaches exist, but the window to act is narrowing.”

Additionally, new research carried out by Professors Rees and Wilsdon, found that the number of grant applications has been increasing in recent years.

In a survey of hundreds of thousands of grant applications from 12 multidisciplinary research funders in six countries that are partners in the Research on Research Institute (RoRI), the funders reported an increase of 17% in application numbers between 2022 and 2024, growing to a 57% increase between 2022 and 2025. This growth ranged from 14% for postdoctoral fellowship applications at the British Academy to 142% for EU Marie Skłodowska-Curie fellowships. There could be several explanations for some of these changes, but the researchers think that AI has played a significant part.

Co-author Professor Wilsdon (UCL Science, Technology, Engineering and Public Policy and Executive Director of the Research on Research Institute), said: “These sharp increases in the volume of funding applications begin soon after the launch of ChatGPT, so it’s likely that a significant portion of this increase is linked to the use of generative AI. This is just the product of earlier versions of large language models: the capabilities of newer agentic systems will drive volumes even higher in 2026.

“Meanwhile, peer reviewers will be using the same agentic tools to assess proposals – so we quickly reach a point where systems of grant funding and review will collapse, unless funders adopt new strategies for managing volume and demand, and for assessing quality.”

However, the researchers caution against clamping down on the use of generative AI by applicants, which would likely be impossible to enforce and inadequate to the challenge at hand. Instead, they urge funders to deploy the power of agentic AI systems to reinvent the funding system, rather than to suppress their use.

This could include using AI to profile applicants from multiple perspectives, allowing funders to identify and compare candidates more completely than a funding panel. It could also include prioritising and shortlisting applications by identifying candidates whose record is consistent with the claims in their application, or by using predictive heuristics that look for novelty and potential impact.

Researchers conclude that when developing these kinds of systems, care is required to avoid reinforcing many of the pitfalls that current funding systems face, such as concentrating resources on those who have already been successful. Transparency would be key to avoid exacerbating biases against early-career researchers, under-represented groups, less established or prestigious institutions, or interdisciplinary and emerging fields.