In a recent article for Nature, Geraint Rees (University College London) and James Wilsdon (UC London) discuss the issues posed when applicants use agentic AI to generate, review, and submit research grants. The authors explain that AI agents offer great productivity gains for those looking to submit research grant proposals, but that this comes at the cost of flooding grant funding organizations with similar applications. In order to avoid a situation in which funders are arbitrarily awarding funding, the authors call for a shift away from written proposals. Instead, they recommend selecting projects for funding base on research performance assessments of the principal investigators and research teams, interviews, and portfolio-based assessments.
