(Vienna, 28-01-2026) A research team from the Medical University of Vienna and the CeMM Research Centre for Molecular Medicine has investigated how so-called AI agents could change the future of biomedical research in a recently published study. The results point to enormous potential for acceleration, but this can only be exploited if it is accompanied by reforms in the exchange and review of scientific findings, investment in shared research infrastructure and targeted support for the introduction of new tools. The study was published in Scientific Reports.
AI systems such as ChatGPT have demonstrated remarkable capabilities in information processing, writing and logical thinking. Building on this, new systems, known as AI agents, are currently being developed that can autonomously plan and execute tasks, use specialised tools and solve complex problems with minimal human guidance. In biomedical research, such agents are already beginning to automate processes such as literature analysis, experiment design and data interpretation. However, it is still unclear to what extent these tools could accelerate research in the future and what obstacles still exist.
Efficiency gains possible in many areas of research
The research team led by Matthias Samwald (Institute of Artificial Intelligence, MedUni Vienna) has now investigated this question. Based on a review of the current literature, the study authors conclude that current AI systems lead to significant efficiency gains in many areas of research. Even more impressive is that recent studies in optimised environments have shown order-of-magnitude speedups for certain activities. These range from automated data analysis and knowledge synthesis to experimental optimisation in robotics laboratories.
The study presents a framework for future AI-supported research, distinguishing between "compressible" and "non-compressible" time. Compressible tasks are those that are primarily based on information processing, such as literature research, data analysis or writing scientific manuscripts. Non-compressible tasks are those that are determined by biological or physical processes and cannot be shortened arbitrarily, even with advanced technology, such as cell division rates, organism development or disease progression. Using a model of a typical multi-year molecular biology project, the authors estimate that the very rapid acceleration of all compressible tasks could speed up progress tenfold. However, biological time constants would ultimately take precedence.
Survey of leading researchers
In addition to the literature analysis, the team conducted an exploratory survey of eight leading biomedical researchers who had authored high-impact publications. The experts reported an average project duration of around six years. While they considered a significant acceleration in structured tasks such as manuscript preparation and administrative processes to be plausible, they were much more sceptical about a drastic acceleration in hypothesis formation and experimental work. A central theme emerged from all the interviews: the ability of the scientific community to adopt new tools and integrate them into existing processes could itself become a limiting factor. This suggests that investment in AI capabilities alone will not be sufficient and that training, shared infrastructure and changes in the organisation and evaluation of research will be equally important.
The authors also noted that AI-driven acceleration of research could exacerbate existing inequalities between well-equipped and underfunded institutions. Access to advanced AI systems and automation infrastructure will increasingly determine which research groups can benefit from these advances. At the same time, the authors point out that realising the full potential of AI agents will likely require a review of established practices. For example, traditional publication and peer review processes, which can extend the timeline of projects by months or even years, may need to be rethought.
"AI agents have the real potential to transform the way research is conducted. However, realising the potential benefits is not only a technological challenge, but also requires us to rethink how we collaborate, publish and review our work, and train the next generation of researchers. Scientific practice will likely need to evolve alongside technology to increase both the speed and accuracy of research," summarises study leader Matthias Samwald.
Publication: Scientific Reports
What are the limits to biomedical research acceleration through general-purpose AI?
Konstantin Hebenstreit, Constantin Convalexius, Stephan Reichl, Stefan Huber, Christoph Bock & Matthias Samwald.
https://www.nature.com/articles/s41598-025-32583-w