MAISON DE LA SIMULATION / Scientific workshop
From assistance to autonomy: agentic AI in research
A day to examine agentic AI in science, its opportunities and limits, and the questions it raises for scientific quality, reproducibility and the organisation of public research.
Workshop — 3 December 2026 — Maison de la Simulation — DIGITEO Building — CEA Saclay
Artificial intelligence has become a routine research tool for data analysis, modelling and programming assistance. Agentic AI represents a further step: systems that can not only analyse information but also participate in the scientific process by exploring the literature, formulating hypotheses, proposing experiments, interpreting results and identifying new research directions. In some cases, intelligent laboratories integrate agentic AI into experimental cycles to coordinate, accelerate and enrich scientific work through hypotheses, simulations and analyses.
Early applications suggest that discovery cycles can be accelerated substantially when AI contributes to several stages of scientific reasoning. This raises major questions for public research: is this an evolution of existing tools or a profound change in research practice? How can scientific quality, reproducibility and control over these systems be maintained? What role can CEA play in a landscape dominated by private technology companies outside Europe?
The workshop will review recent advances in agentic AI for science, identify the opportunities and limitations of these approaches, and begin a collective discussion of strategic directions that CEA could pursue in this emerging field.
Presentations by CEA researchers and colleagues from major international institutions in the United States and the United Kingdom will illustrate recent advances in agentic AI for research and initial work within CEA. The discussion will also examine how agentic AI may transform the research process and its organisation. The day will close with an exchange on practical prospects for deploying these technologies in support of research at CEA.
Christophe CALVIN1 – CEA ADAC – christophe.calvin@cea.fr
Guigone CAMUS1 — CEA Scientific and Technical Data Policy Officer — guigone.camus@cea.fr
1 CEA — Fundamental Research Division — Maison de la Simulation
Programme :
| Start | End | Title | Speaker |
| 09:00 | 09:30 | Welcome | |
| 09:30 | 09:45 | Introduction | C. Calvin (CEA/DRF – FR) |
| 09:45 | 10:30 | AI-Based Scientific Hypothesis Generation: Lessons Learned from Generating 100 Hypotheses per Scientific Problems | F. Cappello (ANL – USA) |
| 10:30 | 11:15 | Agents for scientific discovery: towards autonomous research at institutional scale | B. Bolliet (Department of Physics, University of Cambridge – UK) |
| 11:15 | 11:30 | Break | |
| 11:30 | 12:15 | AI and work: what are the implications for research work? | F. Barcellini (Professor of Ergonomics at Cnam and Director of the Research Centre on Work and Development – CNAM – FR) |
| 12:15 | 13:45 | Lunch | |
| 13:45 | 14:15 | Towards contextualised bioinformatics agents: a traceable knowledge graph for interpreting gene signatures | G. Uguzzoni (CEA/DRF/IRIG – FR) |
| 14:15 | 14:45 | Computational chemistry and energetic materials | C. Wespiser (CEA/DAM/DIF – FR) |
| 14:45 | 15:00 | Break | |
| 15:00 | 15:30 | AI experiments for software development and circuit design | M. Duranton (CEA/DRT/LIST – FR) |
| 15:30 | 16:30 | Round table — Agentic AI for science: a threat, an opportunity or a necessity for research at CEA? |