Arco Bast studies how neurons communicate. Earlier this year, the Janelia postdoc encountered a more mechanical version of the same problem: the components of his custom microscope could not easily communicate with one another. 
The microscope could image thousands of neurons at once, but its cameras, scanners, sensors, and other components relied on different control systems. Coordinating them for a new experiment could require months of custom engineering. 
“It was this feeling of frustration standing in front of this rig,” recalls Bast. “Technically, I could do the experiment that I want to do, but then — because controlling all these heterogenous components is so difficult — I couldn’t really get to my desired experiment quickly.” 
That frustration led Bast to use AI to develop a way for all the parts of his microscope to talk to each other and with AI systems. The fix dramatically cut his setup time, and it could change how millions of scientists work. In collaboration with Anthropicexternal link, opens in a new tab, the developers of Claude, Bast and his Janelia colleagues developed the Model Hardware Standard (MHS)external link, opens in a new tab, which will enable scientists and engineers across the globe to control elaborate hardware systems with AI. 
HHMI Vice President and Janelia Executive Director Nelson Spruston says the innovation will allow scientists everywhere to perform experiments that were never before possible, which could enable new biological discoveries. 
“Arco didn’t just come and do experiments in the same way other people in the lab were doing them at the time — he saw opportunities to do something new and different and went after it,” Spruston says. “Seeing people deploy AI in unique ways in the research environment is really exciting and Arco’s done some great work to make that possible.” 
Bast trained as a medical doctor in Germany. At each step of his training, he wanted to learn more about the mechanisms driving biological processes. He became intrigued by the idea of using mathematical models to explain complex biological systems, leading him to switch his focus to neuroscience.  
“We see there is activity but the real question that drives me is how does it come about mechanistically? What are the components in the brain that enable this type of activity?” Bast says.  
After studying individual neurons in his doctoral work, Bast came to the Spruston Labexternal link, opens in a new tab to study how populations of neurons work together in the brain to form memories. To do that, he needed to track the activity of individual cells as an animal learns.   
Answering novel questions in science often requires custom instrumentation, which means months of trial-and-error getting components that run on different software in separate programming languages to work together. Even with the experts at Janelia, integrating all the pieces can take years.  
At Janelia, Bast had access to state-of-the-art instruments like the mesoscope, but he wanted to use it in a new way, and that would take months of calibration to get it to work just right.  
“I realized that the mesoscope itself is an amazing microscope that, in principle, would allow me to do the many things that are well suited to measure what the cells are doing but not with the control it had,” Bast says. “I was then motivated to deeply think about computers and what they’re doing and why they can’t do the thing that is intellectually pretty simple but takes me so much time.”  
 
Bast took his frustration to Selden Island, the Janelia campus’s 400-acre wooded island on the Potomac River. Away from the hustle and bustle of the lab, he realized that the bottleneck was caused by memory: each mesoscope component was controlled by different software with its own private memory. To communicate with each other, messages needed to pass through a central operating system, slowing down and complicating the process. He reasoned that if the different components could instead share a pool of memory that each could read and write from directly, they could work together.  
Using Claude Code, Bast developed a way for the instruments to do just that, writing an application that enabled all the parts of his experiment to look at and update the same live picture, allowing the system to respond to changes immediately.  
By simplifying communication between the components, Bast cut his setup time from months to days, allowing him to try new experiments, quickly change direction, and focus on his research.  
“The time it takes to set up the experiment is just so much faster,” Bast says. “And that frees up my time to think about science.” 
Bast’s innovation also allows scientists to incorporate AI directly into their experiments. The shared memory allows an AI agent to observe the experimental data being collected, notice patterns, and adjust experiments in real time, tasks that would be impossible for a human to do.  
“Once the rig works, Claude can now, while the experiment is running, look at what’s happening and steer the experiment,” says Bast, who is currently working to implement this idea in his own research.  
Boaz Mohar, a senior scientist in the Spruston Lab who contributed to the development of MHS, says Bast’s creation changed what he thinks is possible in a lab.  
“It means that I can do more projects, but I can also aspire to do more complicated projects,” Mohar says. “It changed what I think is possible as a scientific project.” 
MHS exemplifies the power of an “AI-in-the-Loop” approach, where scientists and AI work together as one continuous system to drive biological discovery. In addition to using AI to speed up and improve scientific experiments, Spruston says AI could also be used to generate new ideas, design ways to test them, and analyze the outcomes — all of which could lead to new scientific discoveries.  
For Spruston, Bast, Mohar, and their colleagues, AI could help them uncover new insights into how brains learn and remember, potentially leading to a better understanding of diseases where these processes go awry, like Alzheimer’s, and the development of better drugs to treat them.  
“We’re just at the tip of the iceberg, in my view,” Spruston says. “What we really want to get to is a state where this integration of AI, along with the experimenter, allows us to do things and see things that we would have never seen before.”  
During a meeting with Anthropic engineers at Janelia in February 2026, Bast explained the problem he had encountered and how he had used AI to overcome it. It became clear that he had made a breakthrough that complemented areas of machine learning that Anthropic was also actively exploring.  
Janelia and Anthropic began a collaboration that led to MHS, a standard for AI agents to safely operate physical devices in scientific research and advanced manufacturing. For scientists, this breakthrough has the potential to be revolutionary, letting them integrate instruments used in unique experimental setups and control them through natural language with AI. 
“With MHS, we’re seeing how AI can remove the barriers between a scientist and their science,” said HHMI President Erin O’Shea. “At Janelia, we’re leveraging AI at every step in the scientific process to accelerate discovery, and Arco’s work is a perfect example of that in action.” 
The next step is for instrument makers to integrate the standard into their hardware, which will enable disparate components to talk to each other from the get-go. In this way, MHS might serve as a universal connector that works for every device regardless of manufacturer, like a USB port.  
“That would be a total game changer for scientists,” Bast says.  
Media Contact: Halea Kerr-Layton, Media Relations Manager [email protected]
Media Contact: Halea Kerr-Layton, Media Relations Manager [email protected]