AI links microscope parts, cutting setup from months to days

Arco Bast was standing in front of a microscope that could image thousands of neurons at once, yet a new experiment still took months to prepare. At the Howard Hughes Medical Institute's Janelia campus, the neuroscientist faced a familiar engineering problem: cameras, scanners and sensors each ran on different control systems. He wanted to study how populations of neurons form memories, but the instrument was not easy to direct in the way his question required.

The bottleneck, Bast concluded, was memory. Each component kept its own private memory, so messages had to travel through a central operating system before the parts could coordinate. Using Claude Code, he built an application that gives the components a shared pool of memory: each can read and update the same live picture of the experiment. The approach became the Model Hardware Standard, developed by Bast, his Janelia colleagues and Anthropic, the company behind Claude.

The immediate result was practical rather than theatrical. Bast says his setup time fell from months to days, giving him more room to try experiments, change direction and concentrate on neuroscience. Boaz Mohar, a senior scientist in the Spruston Lab who helped develop MHS, said the system changed which projects he considered possible, including the possibility of more complicated ones.

The design also creates a path toward an AI-in-the-loop laboratory—an arrangement in which an AI observes an experiment continuously and works alongside the scientist. Because the AI can access the shared live data, it could notice patterns and adjust the experiment while it runs, a task a human could not perform at the same speed. Bast is working to implement that use in his own research; the source does not present it as already operating routinely.

So what changes, concretely? Researchers with elaborate custom instruments could spend less time making incompatible hardware and software communicate, and more time testing biological questions. The benefit is clearest for labs building unusual experiments, where integration can otherwise take months or years. For now, MHS is a Janelia development, and no wider deployment or independent assessment is reported.