AI agents operate scientific instruments and robotic stations through on-the-job learning

1/ AI utilization method: multi-agent LLMs and in-context learning
2/ Applied task: operation of an X-ray nanoprobe beamline and an autonomous robotic station for materials design
3/ Reported result: evaluated the possibility of orchestrating complex measurement workflows and incorporating expert feedback through on-the-job learning
4/ The 2026-09-11 npj Computational Materials study used multi-agent large language models and in-context learning to address the problem of operating an X-ray nanoprobe beamline and an autonomous robotic station for materials design.
5/ The study evaluated whether complex measurement workflows could be orchestrated and whether on-the-job learning could be achieved by incorporating expert feedback.
6/ This case shows that AI agents can actually be used to coordinate multiple stages of scientific instrument operation, but performance and safety requirements change when the instrument type and operating procedures differ.
7/ Therefore, this result cannot be taken as a general automation success, and human supervision and instrument-specific validation are required.
8/ The significance for materials research is that agents can assist tasks with many steps, such as complex measurement procedures or robotic workflows.
9/ However, if measurement failures, data omissions, and tool-call errors accumulate, the entire workflow can proceed incorrectly, so it is important to establish stepwise permissions and intermediate checkpoints. [1]