Astronomers at the National Astronomical Observatories of the Chinese Academy of Sciences (NAOC) are testing artificial intelligence agents that can help plan and carry out telescope observations, an effort that has supported follow-up observations of two early supernova candidates.

The system, known as StarWhisper, was jointly developed by NAOC’s Galactic Three-Dimensional Structure Group and Xinglong Observatory. It combines scientific priorities, weather forecasts and real-time telescope information to help determine what to observe and when.

Using public data from multiple sky surveys, the system has identified and issued alerts for eight very early supernova candidates. Two were followed up by observations when weather conditions allowed, according to the research team.

StarWhisper is part of NAOC’s broader AI for Astronomy programme, which is exploring how artificial intelligence can be used not only to analyse astronomical data but also to support the operation of scientific instruments.

“AI’s role in research is no longer limited to assisting with analysis; it is entering the research execution process itself,” said Dr. LI Yuyang, an expert with the NAOC Artificial Intelligence Steering Committee. “This can allow researchers to focus more on formulating questions and setting objectives, while AI agents support observation execution, validation and iteration.”

The team has built a simulated environment for research-grade telescopes, where AI agents can test observing plans and control strategies before they are applied to real instruments. By modelling equipment status, sensor readings, environmental changes and observation workflows, the environment supports agent training and procedure refinement without using valuable telescope time.

The system has been connected to the Sitian Pathfinder and the Sitian prototype. After researchers submit an observing request, the agent can generate a plan, call relevant telescope-control modules and report results. It can then revise later plans using execution results and expert feedback.

Related work on the StarWhisper Telescope was included as a representative 2025 case in the Science chapter of Stanford University’s AI Index Report 2026, which highlights applications of AI agents across scientific fields.

More broadly, the project provides a testbed for AI-enabled astronomy, in which researchers set scientific objectives and assess scientific results, while AI agents support observation planning, execution and iterative improvement.

NAOC plans to release related models, tools and application capabilities in stages through the CAS’s Panshi large-model platform, supporting their sharing and reuse across the astronomy research community.