A new research framework lets power grid engineers run complex distribution analyses just by typing in plain English.
Researchers built Grid-Orch, a system that connects large language models to OpenDSS, a widely used power system simulation tool, through the Model Context Protocol. It ships with 36 domain-specific tools spanning power flow, voltage analysis, and quasi-static time series simulation, plus three built-in workflows for tasks like capacitor placement and overvoltage mitigation. The system works with cloud models such as Gemini and Claude or with locally-run models like Ollama, so utilities worried about data security can keep everything air-gapped. In one demonstration, a distributed energy resource interconnection screening that normally takes hours of manual scripting finished in under two minutes, producing results identical to direct OpenDSS scripting.
The timing is not subtle. The industry is bracing for a shortage of up to 1.5 million distribution engineers by 2030, driven by retirements and a surge in solar, EV charging, and battery storage hookups that all require grid analysis. Tools like this do not add engineers to the workforce, but they let the engineers who remain move through routine screening work faster, which matters more as interconnection queues back up nationwide.
It's also a sign that the LLM-agent pattern reshaping coding and customer support is creeping into infrastructure engineering, and matching legacy simulation output exactly, rather than approximating it, is the low bar automation has to clear before any utility lets it near a real feeder.