THERE’S A MOMENT in every PCB design project that nobody talks about. You’ve spent days or weeks in your usual EDA software, placing components, defining constraints, getting the stackup right. The board is yours. You know every net, every power domain, every tricky diff pair. Then you export it, upload it to an autorouter, and watch everything you know about your design disappear.
The router sees a bag of components and connections. It doesn’t know that the two nets you labelled USB_P and USB_N need to be routed as a diff pair. It doesn’t know that you deliberately left room near the antenna for a keepout zone. It sees geometry and tries to connect the dots.
This is the gap that has defined automated PCB routing for decades. The engineer holds the intent. The autorouter holds the algorithm. And between them sits a wall of lost context. This is the problem the AI PCB routing assistant is designed to solve.
The DeepPCB assistant is the first step… a collaborator who speaks both the language of design intent and the language of the routing engine.
Alain-Sam Cohen, Head of DeepPCB
It’s a pattern our research engineer, Rick Gentry, kept seeing in the DeepPCB support queue. “We deal with users who try to set up the tool and route their boards, and a lot of the time, the issues they were having were quite simple for us to solve,” he says. “Once we configured the board in a way they hadn’t figured out how to do, it would run and give them a much better result. We wanted to provide someone who knows how to do this to the user, upfront.”
That someone, or something, is the DeepPCB assistant, an AI PCB routing assistant and conversational AI layer now available to all DeepPCB web app users. When a board is uploaded, the assistant begins its own analysis immediately, reading the netlist, identifying components, mapping layers, and flagging what’s placed and what isn’t.
In the PCB design world, assistants aren’t new in themselves; it’s this assistant’s intervention point that’s exciting. Other tools on the market apply AI at the design conception stage. The DeepPCB assistant takes over once those decisions are settled, ensuring the design is configured so the routing agent can actually use it.
The assistant is like a second pair of eyes. It can tell the engineer what’s likely to cause a problem before the routing run and apply a correction across an entire net class with a single command. “Before, you put a board in, you got a board out, and if it didn’t work, you didn’t know why,” Gentry says. “Now, the assistant tells you why it’s failing and how to fix it. That’s the shift from just ‘automation’ to ‘assistance’.”
But the assistant in its current form is only the beginning. The goal is to inject that electrical engineering information fully into the routing and placement agents to make them smarter overall.
In practice, that means moving from guidance to intervention. Rather than advising an engineer to rip up and re-route a poorly connected section of board, the assistant would identify the problem itself, make the change, and hand it back to the routing agent to fix. The DeepPCB assistant is the first step toward this. A collaborator who speaks both languages: the language of design intent and the language of the routing engine. One that can listen when you explain what matters about your board, and act on it. That’s the shift. Not faster routing. Not smarter algorithms. A conversation.
