Human execution errors often mean rework, dragging projects out and burning patience along the way, and the feeling is that more AI involvement would be a good idea.

Quilter reckons the AI-driven workflow slashed what would usually take about three months of human graft down to one week, with just 38.5 hours of expert human involvement. That compares with roughly 430 hours for a skilled engineer doing it the old-fashioned way, which rather underlines where the pain usually sits.

The firm says traditional PCB design gets bogged down in a three-stage slog of setup, execution and cleanup, with execution acting as the real time sink.

Project Speedrun offloaded most of the execution work onto the AI, leaving humans to handle the creative setup and the tidy-up at the end.

Quilter says its system can handle all three stages on its own if asked, but the sweet spot is to let humans steer while the AI handles the repetitive heavy lifting.

Venture Beat said that Quilter’s AI is nothing like language models such as GPT-5 or Claude, because circuit boards were not a word game.
Instead, the system was trained by playing an optimisation game directly against the laws of physics.

Quilter did not train the AI on human-designed boards because humans make mistakes, and the company did not want its AI to be constrained by human-level design habits.

The gamble seems to have paid off, at least this time, with one engineer reacting to the first successful boot by saying, “Holy crap, it’s working.”

Quilter chief executive Sergiy Nesterenko said the long-term ambition is to “come up with better designs for circuit boards than humans have ever tried to do."

The startup is pitching this as a way to lower the barrier to entry for new hardware companies by stripping out one of the most expensive and time-consuming steps.