A team of researchers has developed an artificial intelligence system made up of several specialist agents that can turn research questions posed in plain language into executed, checked and reproducible quantum simulations. The system co-ordinates multiple specialist software packages to carry out the work end to end, without the user having to write code.

The approach is motivated by the growing complexity of the quantum simulation workflow: choosing the right package, writing the scripts, checking the results and ensuring that another researcher can repeat the calculation are time-consuming and error-prone tasks. Splitting those functions among AI agents allows each step to be delegated to a component designed for it, from interpreting the question to validating the output.

The system translates the plain-language query into a simulation plan, runs it with several widely used quantum software packages and subjects the results to automatic checks before delivering them. The verification step and the logging of the procedure are intended to make the result not only correct but also reproducible by third parties, a standard requirement in computational research.

The authors present the work as a way to bring quantum simulation tools within reach of researchers who are not computing specialists, and to reduce the operational burden on those who are. Open questions include the reliability of the automatic checks, the risk that the agents introduce silent errors, and the need for the community to evaluate the system on real research problems beyond the initial test cases.