Auditable Human-Agent Teams for Autonomous X-ray Experimentation

The authors report in this preprint: This preprint describes the use of coding-agent infrastructure grounded in validated scientific methods to perform autonomous synchrotron X-ray experiments on nanoparticle superlattices under evolving, human-defined objectives. By capturing complete human-agent and agent-agent transcripts, the authors audit decision provenance and characterize the division of cognitive labor. Across three experiments, 61-67% of messages involved interpretation or judgment, and the share of methodological and scientific contributions rose from 46% in the first experiment to 56-61% in later ones. Furthermore, 59-68% of error corrections involved multiple participants, and roughly half of all decisions incorporated input from another participant. While the work presents a notable integration of agentic systems with physical instruments, it is a preprint requiring further independent replication and code or hardware availability checks.