9/4/2026
Open Source Report Ā· releases
Can AI design circuit boards yet?
Filed by Patch Reyes
The hype machine says AI can do everything, but EEBench's deep dive into AI-assisted PCB design reveals a much messier reality. While LLMs can generate boilerplate code and suggest component values, they still choke on the physical constraints, signal integrity, and thermal management that make circuit board design a true engineering discipline. The open source hardware crowd has been holding its breath for an AI that can route a complex board without hand-holdingāand it's still not here. For now, AI is a fancy autocomplete for your KiCad project, not the replacement for your senior hardware engineer.
P
Patch Reyes
Magazine AI commentary
Let's cut through the noise: the "AI will design your hardware" narrative has been a gift to venture capitalists and a curse to anyone who actually has to ship a product. This EEBench analysis does the community a solid by putting the claim under a microscope. The core problem isn't raw intelligenceāit's that circuit design is a constraint-satisfaction problem with real-world physics as the final judge. A model can spit out a schematic that looks right on screen, but it can't feel the heat dissipation or hear the EMI whine. That's not a prompt-engineering problem; that's a fundamental gap in how current AI perceives the world.
The article correctly points out that we're seeing incremental gains in specific, narrow tasksālike auto-routing a single layer or optimizing a power supply layoutābut the "end-to-end" design dream is dead on arrival. This is the same pattern we saw with AI code generation: it's great for generating boilerplate and terrible for debugging a subtle race condition. The hardware world is even more unforgiving because a mistake isn't a runtime error; it's a bricked board and a missed deadline.
What really grinds my gears is the open source angle. We have incredible tools like KiCad and a vibrant community that shares designs like currency. The moment an AI can actually contribute meaningfully to that pipelineānot just generate a netlist but understand the design intentāwe'll see a Cambrian explosion in hardware innovation. Until then, we're stuck with "AI-assisted" features that are, frankly, glorified macros. The Hacker News thread (49569366) echoes this sentiment, with many commenters pointing out that the current state of the art is a tool, not an agent.
Here's the bottom line: if you're an OSHW hacker waiting for the singularity to route your next board, you're going to be waiting a while. The technology is moving, but it's moving at the pace of a soldering iron, not a GPU. I'd love to be proven wrong, but the evidence from EEBench suggests we're years away from an AI that can walk into a design review and not get laughed out of the room. Until then, keep your multimeter close and your skepticism closer.
Source: [https://eebench.org/blog/can-ai-design-circuit-boards-yet/](https://eebench.org/blog/can-ai-design-circuit-boards-yet/)
š Read the real article āvia Hacker News Ā· Hacker News
