There's a reason AI got so good at software so fast, and that same reason is why you can't fully trust it with hardware.
When AI writes software, it can run the code, see what breaks, and try again, as many times as it takes.
Hardware makes that feedback much harder to get. You can simulate circuits, but eventually someone has to build the product and test it.
That means paying for boards, waiting for them to arrive, and finding out what actually works.
A member of my Hardware Academy ran into this while developing a wireless tracker with very little room to spare.
He'd been working with Claude, which at first confidently told him the design couldn't be done. Then he gave it the chips he had in mind from his own research, and Claude reversed itself and produced a complete PCB layout.
It was every bit as confident the second time as the first.
To his credit, he knew he didn't have the experience to judge the design himself. So he shared it with us and asked for help finishing it.
Within a day of expert review, two problems surfaced that Claude hadn't flagged.
One of the core modules had the wrong footprint. The actual part was larger, and in a design that tight, that's fatal.
The radio chip in the design would also put the product on an unnecessarily long, expensive certification path.
He fixed both before a single board was ordered.
Neither problem required building his board to catch it. But the experience behind that review comes from working on real products.
An experienced hardware engineer has already been through that build-test-fix cycle on other designs. They've tested boards that didn't work as expected, tracked down the cause, and changed the design.
They've dealt with problems that showed up during certification or manufacturing.
Those lessons carry over to the next product. They change what the engineer questions and which details make them stop and take a closer look.
You get the benefit of that experience before spending money on your own boards.
AI can draw on engineering knowledge, including documented failures. But it doesn't have that same firsthand history of taking products through development and manufacturing, then having to deal with what went wrong.
And when it hands you a design, how do you decide whether to trust it?
That's the same problem you face when you hire an engineer to develop your product, or if you've learned to design the board yourself.
You can't judge the quality of work you don't understand. And on your first design, you can't see what you don't know to look for.
With software that's survivable, at least at first, since even a non-programmer can run the app and see whether it works.
But even there, the free test only goes so far. Whether that app holds up once it's deployed and scaled is something only an experienced software engineer can judge.
Hardware hits that wall on day one since there's no free test it can run.
So whether the design came from AI, from someone you hired, or from your own hands, the only check is the judgment of an engineer who understands it.
This member recognized that. He knew a finished-looking layout isn't the same as a finished design, so he brought it to people who could evaluate it.
That's what the experts in the Academy do every week. Their experience helps you catch problems before you have to learn about them the expensive way.
Get expert eyes on your product inside the Hardware Academy.
Talk soon,
John Teel
Predictable Designs
|
|
|
|
|
|
|
