Agentic AI materials discovery is emerging as a new approach to research automation that extends beyond data analysis to include planning, execution, and iterative experimentation. A newly announced initiative led by Lawrence Berkeley National Laboratory aims to operationalize this concept for energy-related materials research, with potential implications for how laboratories manage computational resources, experimental throughput, and data validation.
The project, known as FORUM-AI (Foundation Models Orchestrating Reasoning Agents to Uncover Materials Advances and Insights), is a multi-institutional effort supported by the US Department of Energy (DOE) through its Scientific Discovery through Advanced Computing (SciDAC) program. The four-year, $10 million initiative seeks to develop an open-source, automated materials discovery platform that connects hypothesis generation, simulation, and laboratory experimentation into a single research pipeline.
What agentic AI materials discovery enables
Agentic AI materials discovery differs from conventional machine learning approaches by allowing AI systems to take action rather than only generate predictions. In practice, this means the system can propose research hypotheses, determine appropriate computational or experimental methods, execute those methods, and evaluate outcomes with limited human intervention.
FORUM-AI is designed to support AI-driven materials research across multiple stages of the discovery process, from computational screening to experimental validation. The platform is intended to orchestrate large numbers of parallel simulations and experiments, reducing the time required to identify promising materials for applications such as batteries, semiconductors, and energy technologies.
Infrastructure supporting AI-driven materials research
The automated materials discovery platform will rely on leadership-class computing and laboratory infrastructure across the national lab system. High-performance computing resources, including the National Energy Research Scientific Computing Center, the Oak Ridge Leadership Computing Facility, and the Argonne Leadership Computing Facility, enable the system to evaluate hundreds of research pathways simultaneously.
On the experimental side, FORUM-AI will integrate with automated synthesis facilities such as Berkeley Lab’s A-Lab, which performs computer-controlled inorganic powder synthesis. Physics-based simulation tools and established computational methods are embedded within the platform to ensure that AI-generated recommendations align with validated scientific models.
Data integrity and transparency in automated workflows
A key concern in AI-driven materials research is ensuring that outputs are accurate and traceable. FORUM-AI addresses this by relying on curated materials databases rather than model memory alone. When queried for specific material properties, the system retrieves values from verified data sources.
The platform also emphasizes transparency. Research plans and reasoning traces generated by the AI are designed to be inspectable, allowing researchers and lab managers to review and modify workflows before allocating resources. This approach supports reproducibility while maintaining human oversight within an automated materials discovery platform.
Operational considerations for lab managers
For laboratory managers, agentic AI materials discovery introduces new operational considerations, including automation readiness, data governance, and workforce training. Managing AI-driven experimentation may require updated protocols for validation, cross-functional collaboration, and system oversight.
The FORUM-AI project reflects a broader organizational shift toward integrating AI, robotics, and high-performance computing as core research infrastructure. As automated materials discovery platforms mature, lab leaders will need to assess how these systems align with existing workflows, safety practices, and long-term research strategy.
What comes next for FORUM-AI
By the end of the project, the FORUM-AI team aims to deliver a fully integrated, end-to-end platform for automated materials discovery. Future development may include connections to additional experimental user facilities, enabling AI systems to assist with experiment preparation or execution.
As agentic AI materials discovery transitions from development to deployment, the initiative offers a practical example of how AI-driven materials research could reshape laboratory operations in the coming years.
This article was created with the assistance of Generative AI and has undergone editorial review before publishing.
