How does an electronics design review work?
It’s an essential process, but it has its limits. We might budget 12 hours to review a PCB, and an independent reviewer first has to take the time to understand the project history and the designer’s intent. Even with clear procedures and a checklist, a review lets some errors through: there’s no such thing as zero risk. We could spend more hours on it, but past a certain point, every extra effort costs more than fixing the board once it’s built.
“Even with clear procedures and a checklist, a review lets some errors through.”
What has changed since you started using AI?
“A lot has changed over the past six months. Six months ago, if you’d asked me to use AI for a design review, I’d have said no way: half the time, the answers were wrong. I was using Google’s Gemini, which at the time worked well for quick, simple questions, but not for this kind of task. Then I switched to Claude, Anthropic’s AI assistant, three or four months ago.
“If you don’t guide the AI in the direction you want, you’ll be disappointed with the results.”
Where does AI fit in the review process?
“When I review another designer’s board, the process stays the same, with one difference: there are now two parallel tracks. I do my own review while the AI does its own. I combine its findings with mine, the designer makes the changes, then the AI and I each do a second review before fabrication. It’s like adding a second independent reviewer.
During design, I bring it in as early as possible. When I’m choosing components, I ask whether it sees options I hadn’t considered. Once a circuit is done, I run it through the AI to spot bad connections or edge cases that could cause errors. Then, once all the circuits are connected, I have it do a global review.
“I make the decisions, not the AI.”
What does AI catch that human miss?
“Sometimes it’s simple things. On one project, the part number shown on the schematic didn’t match the one in the bill of materials (BOM). We would have installed the wrong part on the board, then lost a lot of time tracking down the problem. The part had already been validated, and two people had reviewed the schematic since. Neither of them noticed. The AI did.
“In electronics, there’s what we call ‘black magic’: problems you only run into during testing. AI finds quite a few of them.”
Where does AI get things wrong?
“Its mistakes mostly come from its sources. It once claimed a component would run at 4 V when the datasheet listed a 3.5 V maximum: it had relied on a generic datasheet for the same part family. Now I give it the datasheets by default, and I ask it not to rely on opinions from the internet without showing me its sources. Just because someone on Reddit says it works doesn’t make it true. And AI tends to sound very sure of itself.”
Where is the real value of AI in design reviews?
There will always be some errors left. But what we’re hoping for with AI is a list that drops from 20 items to two, and a respin that takes 20 hours instead of 60.
We usually build several prototypes for testing and for our embedded software team. A manual fix, repeated on 5 or 10 boards, adds up fast. Not to mention that an error caught early can save us a board respin. In the end, this in-depth AI review is an investment.
“The AI flagged some very in-depth issues. The review took longer than we’d planned, but we’ll more than make up for it later.”
Looking ahead, Daniel hopes to see AI built directly into design software, which would spare him long rounds of feeding it context. In the meantime, he’s sure of one thing: in 2026, AI has become an essential ally in his design reviews. It doesn’t replace his best practices or his engineering judgment, but it adds one more layer of verification.