The Agentic Revolution: Double Bill on Agentic Workflows
Talk details
Date
May 20, 2026
Time
11:00am PST / 2:00pm EST / 20:00 CEST
Overview
The (R)evolution of Scientific Workflows in the Agentic AI era: Towards Autonomous Science
Woong Shin (Oak Ridge National Laboratory)
Modern scientific discovery increasingly requires coordinating distributed facilities and heterogeneous resources, forcing researchers to act as manual workflow coordinators rather than scientists. Advances in AI leading to AI agents show exciting new opportunities that can accelerate scientific discovery by providing intelligence as a component in the ecosystem. However, it is unclear how this new capability would materialize and integrate in the real world. To address this, we propose a conceptual framework where workflows evolve along two dimensions, intelligence (from static to intelligent) and composition (from single to swarm), to chart an evolutionary path from current workflow management systems to fully autonomous, distributed scientific laboratories. By embedding reasoning and adaptation into workflows, these labs have the potential to accelerate discovery by factors of 10 to 100, transforming exploratory science into a continuous, machine-augmented process.
Large Language Model Agents for Atomistic Simulation Workflows
Jan Janssen (Max Planck Institute for Sustainable Materials)
Large language models (LLMs) are powerful tools for scientific research, particularly in interdisciplinary fields such as materials science, where they can bridge the gap between theory and experimentation. We have explored the potential of LLMs in designing new materials through atomistic simulations. Although LLMs face challenges such as complex Python programming and generating accurate responses, these issues could be overcome by integrating an agentic approach with simulation workflows in the pyiron framework. Our pyiron framework allows LLMs to calculate material properties and perform inverse design, identifying alloy compositions that meet specific criteria. The study highlights the importance of linking research data to scientific workflows to simplify processes for LLMs and researchers alike. The LangSim interface, which combines language and simulation, has been developed to facilitate this integration and is available on GitHub, thereby enhancing the synergy between LLMs and scientific workflows in materials science.
Woong Shin (Oak Ridge National Laboratory)
Modern scientific discovery increasingly requires coordinating distributed facilities and heterogeneous resources, forcing researchers to act as manual workflow coordinators rather than scientists. Advances in AI leading to AI agents show exciting new opportunities that can accelerate scientific discovery by providing intelligence as a component in the ecosystem. However, it is unclear how this new capability would materialize and integrate in the real world. To address this, we propose a conceptual framework where workflows evolve along two dimensions, intelligence (from static to intelligent) and composition (from single to swarm), to chart an evolutionary path from current workflow management systems to fully autonomous, distributed scientific laboratories. By embedding reasoning and adaptation into workflows, these labs have the potential to accelerate discovery by factors of 10 to 100, transforming exploratory science into a continuous, machine-augmented process.
Large Language Model Agents for Atomistic Simulation Workflows
Jan Janssen (Max Planck Institute for Sustainable Materials)
Large language models (LLMs) are powerful tools for scientific research, particularly in interdisciplinary fields such as materials science, where they can bridge the gap between theory and experimentation. We have explored the potential of LLMs in designing new materials through atomistic simulations. Although LLMs face challenges such as complex Python programming and generating accurate responses, these issues could be overcome by integrating an agentic approach with simulation workflows in the pyiron framework. Our pyiron framework allows LLMs to calculate material properties and perform inverse design, identifying alloy compositions that meet specific criteria. The study highlights the importance of linking research data to scientific workflows to simplify processes for LLMs and researchers alike. The LangSim interface, which combines language and simulation, has been developed to facilitate this integration and is available on GitHub, thereby enhancing the synergy between LLMs and scientific workflows in materials science.