Artificial intelligence is shifting from a tool that helps scientists work faster to an autonomous agent that runs experiments independently, monitors results in real time, and steers its own discovery process. Researchers at Oak Ridge National Laboratory (ORNL) showcased this transformation at the 2026 AI+ Expo for National Competitiveness in May, demonstrating how AI agents integrated with supercomputers and laboratory equipment are accelerating materials discovery and manufacturing across multiple domains.

What Are AI Agents Doing in the Lab Right Now?

ORNL researchers presented three distinct applications of autonomous AI systems that illustrate how the technology is moving beyond data analysis into active experimental control. Each system combines artificial intelligence with high-performance computing and specialized laboratory instruments to automate tasks that traditionally required constant human oversight.

• Metal 3D Printing: A system called LOOP uses AI agents and supercomputing to monitor metal parts during manufacturing in real time, automatically adjusting the printing process to ensure reliability for critical energy infrastructure components.
• Quantum Materials Research: AI and advanced computing methods analyze experimental data from neutron sources faster and more accurately, uncovering hidden spin interactions that drive quantum material behaviors and accelerating discovery of materials for advanced electronics and quantum computing.
• Plant Phenotyping: The Orchestrated Platform for Autonomous Laboratories (OPAL) combines multiple imaging instruments, exascale computing on the Frontier supercomputer, and autonomous decision-making to predict plant traits across large populations in sub-minute response times, collapsing week-long detection cycles into near-instantaneous analysis.

The shift toward autonomous laboratories represents a fundamental change in how scientific discovery works. Rather than scientists manually collecting data and running analyses, AI agents now coordinate instruments, process results on supercomputers, and decide what experiments to run next, all within a continuous feedback loop.

How Is This Different From Traditional AI in Science?

The key distinction lies in agency and autonomy. Traditional AI in research acts as a research assistant, helping scientists analyze data or identify patterns after experiments are complete. The new generation of AI agents at ORNL actively participates in experimental design and execution, making decisions about what to measure, how to adjust parameters, and which hypotheses to test next.

"At ORNL's Advanced Plant Phenotyping Laboratory, we are building a specialized set of AI agents to coordinate imaging instruments, exascale inference on Frontier, and experimental decision-making inside a single closed loop. We're moving past the era of AI as a research assistant and into one where agents actively run the science. Coupling foundation vision models with exascale-class agentic workflows turns high-throughput plant phenotyping into a discovery engine that learns continuously and steers itself, with scientists on the loop where it matters most," explained Rafael Ferreira da Silva, distinguished research scientist at ORNL.

Rafael Ferreira da Silva, Distinguished Research Scientist at ORNL

This approach maintains human oversight at critical decision points while eliminating the bottleneck of manual data collection and preliminary analysis. Scientists remain involved where their expertise matters most, such as interpreting unexpected results or setting long-term research goals, rather than spending weeks processing routine measurements.

Why Does This Matter for Materials Science and Energy?

The acceleration of discovery cycles has direct implications for developing new materials for energy infrastructure, quantum computing, and advanced electronics. By collapsing week-long analysis periods into minutes or seconds, researchers can test more hypotheses, explore larger design spaces, and identify promising materials faster than traditional methods allow.

ORNL's demonstrations occurred within the context of the Department of Energy's Genesis Mission, a national initiative aimed at doubling the productivity and impact of American science, engineering, and research and development within a decade. The mission integrates AI, high-performance computing, advanced scientific instrumentation, and multidisciplinary expertise across the DOE ecosystem.

"The Genesis Mission reflects a bold national commitment to accelerate discovery by integrating AI, high-performance computing, advanced scientific instrumentation, and multidisciplinary expertise across the Department of Energy ecosystem," stated Gina Tourassi, ORNL Associate Laboratory Director for the Computing and Computational Sciences Directorate.

Gina Tourassi, ORNL Associate Laboratory Director for the Computing and Computational Sciences Directorate

The American Science Cloud, a cornerstone initiative within the Genesis Mission, aims to provide researchers across the national laboratory system with shared access to AI capabilities and computing resources. This infrastructure enables smaller research groups to leverage the same autonomous laboratory technologies that ORNL has developed, democratizing access to advanced discovery tools.

How to Implement Autonomous Laboratory Workflows in Your Research

While ORNL's systems represent cutting-edge implementations, the underlying principles can guide other institutions and research groups considering autonomous laboratory approaches:

• Start with High-Throughput Processes: Identify experiments that generate large volumes of data or require frequent parameter adjustments, such as materials screening, phenotyping, or manufacturing quality control, where automation provides the greatest time savings.
• Integrate Specialized AI Models: Deploy foundation vision models or domain-specific machine learning models that can interpret experimental data in real time, enabling the system to make informed decisions about next steps without human intervention.
• Connect to High-Performance Computing: Link laboratory instruments to supercomputing resources that can process complex analyses at scale, allowing AI agents to handle computationally intensive tasks like quantum simulations or large-scale image processing within experimental feedback loops.
• Maintain Human Oversight at Decision Points: Design autonomous workflows that flag unexpected results, novel findings, or major experimental decisions for human review, ensuring that scientists remain engaged in interpreting surprising outcomes and setting research direction.

The convergence of AI agents, supercomputing, and laboratory automation represents a new paradigm in scientific discovery. Rather than viewing AI as a tool that augments human researchers, ORNL's approach treats AI as an active participant in the scientific process, capable of running experiments, analyzing results, and proposing next steps with minimal human intervention. This shift has the potential to accelerate the pace of materials discovery and energy innovation across the national laboratory system and beyond.