Every electrical engineer knows the drill. The mechanical model shows up, and somewhere in there are dozens of pieces of equipment waiting to be connected. Heat pumps, air handlers, unit heaters. Each one needs a circuit, a panel with available capacity, and a feeder that can handle the load. It’s necessary work, but it’s not the work you went to engineering school for.
So we decided to see what would happen if we handed that work to AI.
The Setup
We started with a real Revit project and a simple question: what’s going on with the mechanical equipment? Claude dug into the model and came back with an inventory of 47 pieces of equipment. The interesting part was at the bottom of the list: 17 of them weren’t connected to the electrical system at all.
Rather than fixing them one at a time, we asked the AI to study the electrical distribution and come up with a circuiting plan.
What the AI Actually Did
This is where it gets interesting. The AI mapped out the entire distribution on its own: the service transformer, the switchboard, the MDPs, and the panels downstream. Then it started building a plan for connecting the equipment.
But it didn’t just run off and make changes. As it worked, it paused to ask questions an engineer would want to weigh in on. How should the heat pumps be split between two panels? What should happen when a panel turns out to be undersized? When it ran out of physical space on a panel, it laid out the options: create a new panel, overflow to another, or leave the connection dangling.
We made the calls. It did the work.
Where the Engineer Still Matters
The AI got all 47 pieces of equipment connected, but it was honest about its limits. It created a new panel to handle the overflow, then flagged that the panel was floating in space and needed a real home in the building. Finding that home led straight to a familiar problem: the room was too small. That’s a conversation with the architect, and no AI is having that conversation for you. Yet.
There was also finishing work on the single-line diagram and a voltage drop issue that still needs an engineer’s attention. This isn’t push-button design. It’s more like handing the repetitive modeling work to a capable assistant who checks in when decisions need judgment.
The Bottom Line
The whole process took the AI 15 to 20 minutes of real time. Could an experienced engineer have done it faster by hand? Maybe. But here’s the thing: while the AI was working through panel schedules and circuit assignments, the engineer was free to work on something else entirely.
Try It Yourself
On the back end, this runs on an MCP server, and it’s currently in early release with interested customers already using it on their projects. If you’d like to be one of them, reach out to us at designmaster.biz, and we’ll get you set up.