You’ve probably tried it by now. You feed a schematic to your favorite LLM, ask it to route a simple 8-bit microcontroller board, and sit back waiting for the future of hardware engineering to arrive.
Instead, you get a disaster. Traces that loop back on themselves. Ground planes that don’t exist. Vias placed exactly where they shouldn’t be. As one frustrated engineer put it recently, AI makes too many “stupid errors” to be trusted with actual layout.
We are obsessed with making AI replace the architect, when its true superpower is doing the janitorial work.
The fundamental problem is that language models are probabilistic, conversational engines. PCB layout is deterministic, multi-constraint spatial routing. You can’t “hallucinate” your way through impedance-controlled routing. Asking an LLM to draw a physical board is like asking a poet to build a bridge. The words might sound beautiful, but the structure will collapse under its own weight.
But here is the twist: while we were all laughing at AI’s spectacular failures in physical execution, it quietly became brilliant at the most agonizing minutiae of hardware design.
Have you ever spent three days trying to source a specific MOSFET from Digikey or LCSC, only to find it’s out of stock, forcing you to dive into 400-page datasheets to find a drop-in replacement? That is the actual pain point of engineering. Not the routing. The drudgery.
An AI won’t draw your board, but it will read 400 pages of Texas Instruments datasheets so you don’t have to.
The real value of AI in hardware isn’t replacing the engineer’s creative mind; it’s eliminating the soul-crushing clerical work that surrounds it. Sorting out component libraries. Double-checking footprints. Managing the BOM. Talking through component choices to coach you on optimization errors.
Engineers are still drawing schematics. They are still laying out boards. The creative, deterministic work remains firmly in human hands. But the tedious edges of the process? The datasheet reading and substitute sourcing? That is being automated right now.
The revolution isn’t artificial engineering; it’s automated clerical work.
We need to recalibrate our expectations. AI isn’t taking over PCB design. It is fundamentally reshaping the workflow by offloading the auxiliary tasks we hate. Let the AI manage the libraries and read the datasheets. Pick up the mouse and route the board yourself.
FAQ
Q: But isn't AI supposed to learn spatial constraints eventually?
A: Maybe, but language models aren't built for deterministic geometry. Routing requires strict rules, not probabilistic guessing. Don't hold your breath for an LLM to replace Altium anytime soon.
Q: What's the practical implication here?
A: Use AI for component sourcing, BOM management, footprint checking, and datasheet parsing. Let it handle the library management while you do the actual design and layout.
Q: Does this mean AI is actually useless for hardware design?
A: No, it means we've been measuring its value wrong. Eliminating datasheet drudgery saves more engineering hours than a mediocre auto-router ever could.