Someone building his first product asked AI to design a small circuit for him.
The answer came back in seconds, with every part labeled and every connection spelled out.
He built it exactly the way AI told him to, and it didn't work.
I use AI every single day in my business, and if you're developing a product right now, you should be using it too, as much as you possibly can.
But there's one line AI can't cross no matter how good the models get.
Knowing where that line sits is the difference between AI saving you weeks and AI costing you months.
It all comes down to the cost of being wrong.
In software a wrong answer is cheap, since you just patch it and move on.
In hardware, a confidently wrong answer can mean a board respin, a failed round of certification testing, or a production order you can't unwind.
And AI sounds exactly as confident when it's wrong as when it's right, so it can never warn you which of its answers landed in the small fraction it made up.
So I'm going to walk you through six hardware decisions you should never trust AI to make, and what to do instead on every one of them.
Click here to watch the video or read the blog
Okay, let's get started.
The rule I follow is simple: use AI where being wrong is cheap, and use experienced engineers and product experts where being wrong is expensive.
On the cheap side of that line, AI is incredible, and I lean on it constantly.
I use it to explain concepts I'm learning, to write first-draft code, and to summarize a 300 page datasheet so I know exactly where I need to slow down and read closely.
It'll surface options fast once I already know the question I'm asking.
I even use it to pressure test my own reasoning, sort of like a colleague I can think out loud with.
Every one of those uses shares the same trait: a wrong answer costs me nothing, since I'll either catch it or iterate right past it.
The six decisions coming up are the opposite.
These are the ones where a wrong answer gets built into hardware, or into a purchase order, or into a certification submission before anybody notices it.
None of this is me telling you AI is bad at engineering, because it isn't.
It's me telling you where the price of a mistake changes.
Decision #1 - Which Components Go Into Your Product
AI is trained on the internet, and the internet is mostly hobbyist content.
So when you ask it to pick parts for you, it hands you the popular hobbyist answer instead of the production-grade one.
That's how an Arduino board ends up sitting inside a production design, or how the module every blogger loves ends up in a product that has to pass certification and ship at volume.
I've seen designs built almost entirely with AI where the footprints didn't match the actual parts, and where several of the components couldn't be sourced at production quantities at all.
One of them had a hobbyist module sitting right in the middle of what was supposed to be a production-ready board.
Production parts get chosen for reasons that never show up in a tutorial.
You're asking whether the manufacturer will still be making that part in five years, whether you can buy ten thousand of them next quarter, and whether there's a pre-certified version that saves you an entire round of testing.
A component choice isn't really a technical question anyway.
It's a sourcing question, a lifecycle question, a cost question, and a certification question all at the same time, and AI has no idea what your volumes, your margins, or your target countries are.
The better way to use it here is backwards: pick the part yourself using production criteria, then hand AI the datasheet and let it explain the sections you don't understand.
Decision #2 - Whether a Circuit Design Actually Works
The circuit I opened this video with was a latch, which is just a small circuit that holds a signal on or off until you tell it to change.
After days of tweaking values and getting nowhere, he went back and asked the AI to rework the whole design.
The second answer came back even more polished, with pin tables and little checkmarks next to every step.
Underneath it was the exact same wrong circuit.
One of my engineers rebuilt the design in a simulator, watched it fail there the same way it had been failing on the bench, and worked out that the product needed a completely different type of latch.
The lesson is that a beautifully formatted wrong answer is still wrong, and it'll be wrong in exactly the same way the second time you ask.
AI doesn't notice its own mistakes, no matter how nicely you word the follow-up.
When you ask it to try again, you're not getting a review.
You're getting a rewrite of the same reasoning with better presentation.
What was missing both times was verification: simulating the circuit, breadboarding it before you commit it to a layout, or putting it in front of somebody who will test it and then stand behind the result.
Get your ideas from anywhere you like, including AI, but never let an untested idea walk straight into your schematic.
Decision #3 - Whether Your Design Is Ready to Build
Plenty of people are now uploading a schematic or a layout and asking AI to review it, and the design tools themselves are shipping AI copilots that do the same thing.
Those tools do catch real problems, and I'm not going to pretend otherwise.
But they miss far more than they catch.
The part that should worry you is that you have no way of knowing what got missed.
An experienced reviewer tells you what they looked at, so when they don't flag your power supply, that silence actually means something.
With AI you get a list of findings and no idea what was covered.
That means a clean report and a report from a tool that skipped half your board look identical to you.
Back at Texas Instruments, every design I worked on had to go through a formal review where a room full of engineers picked it apart in excruciating detail.
I was terrified of those meetings for my first couple of years.
But that process existed for a reason, since designing in isolation invites problems no matter how experienced you are.
That's just as true on your hundredth design as it is on your first.
So the right way to use AI here is as a first pass that clears out the obvious stuff, like a missing decoupling capacitor or an unconnected net, before a person looks hard at the parts that actually cost money to get wrong.
Decision #4 - Whether That Cheaper Part Is Actually Cheaper
Cost decisions are where AI sounds the most reasonable and helps the least, since it's reasoning from generic knowledge instead of your actual numbers.
The classic version of this is a wireless design, where the bare chip runs a couple of dollars cheaper than the pre-certified module.
On a spreadsheet, that looks like an easy win.
What that swap actually buys you is full FCC and CE testing on your own design, antenna tuning that's now entirely your problem, and a retest every time your layout or your enclosure changes.
Ask AI to weigh that tradeoff and it'll give you a perfectly sensible sounding answer.
It doesn't know your volume, your margins, your target retail price, or which countries you're selling into.
The same thing happens with the part that's technically fine but wrecks your margins, or the cheaper option that sails through your design and then fails a safety test six months later.
Certification costs also don't scale the way component costs do.
Saving two dollars a unit means nothing at all if it triggers another round of lab testing that costs more than your entire first production run.
AI can still help you with cost tradeoffs, but only if you give it your real numbers, your real volumes, and your real target markets.
Just remember that AI doesn't carry your full picture from one conversation into the next, at least not the way a person working with you does.
The chat where you carefully laid out your volumes and your margins is not the chat where you come back two weeks later and ask it to adjust your design.
Unless you re-supply that entire picture every single time, it falls back to generic reasoning, and it will never tell you that's what happened.
An experienced engineer carries your whole project around in their head and brings it into every conversation you have with them.
So treat whatever AI hands you as a starting point that somebody who's been through certification still needs to check.
Decision #5 - Whether Your Product Idea Is Any Good
The single worst question you can ask AI is whether your idea is a good one.
AI is the most agreeable advisor you will ever have, and it's biased toward telling you yes.
A yes you went looking for works a lot more like a mirror than a market check.
That's the same yes you get from friends and family who want to see you win, and it's worth about the same amount.
I learned this one the expensive way with my own product.
I treated trade show compliments and a single retail store test as if they were proof that people would actually buy it.
Interest isn't validation, enthusiasm isn't either, and an AI telling you the market opportunity looks promising is worth less than both of those.
Validation is somebody handing you money, or a store reordering because the first batch actually sold through.
There's no shortcut around that.
The trap is that a confident yes feels like progress, so you keep building instead of going out and finding the one person willing to pay you.
If you're going to put your idea in front of AI at all, flip the prompt around and ask it to argue against you, list every reason this fails, and tell you what would have to be true for it to work.
Even then, treat that output as a list of things to go verify with actual customers, not as an answer.
Decision #6 - What You Should Be Asking In The First Place
Everything so far has been about wrong answers or flattering ones.
The biggest failure is a different thing entirely, because it's about the questions that never get asked at all.
AI is at its best when you already know exactly what your problem is, and it gets much weaker the moment you're not sure what the problem even is.
On its own, it will never raise a question you didn't think to ask.
I've leaned on AI hard in my own business for years, and it's given me real help and plenty of good suggestions.
But not once has it surfaced a blind spot I'd been operating with for months, since I had to find the question myself before it could help me answer it.
In hardware those unasked questions get expensive fast, like the certification you never knew applied to your product, or the manufacturing constraint you never thought to raise with your factory.
The sneakiest one is the design where every individual answer was correct, and the whole thing still falls apart the moment the subsystems have to work together.
AI answers the prompt in front of it, while an experienced reviewer looks at your entire project and says wait, this won't work.
So when you hit one of those expensive decisions, the ones where you don't know what you don't know, you really only have three options.
You can work through it yourself, which is fine when you know what to look for.
You can hire an independent engineer who has actually taken a product through manufacturing and certification, or you can get in front of people who have already made the mistake you're about to make.
Click here to watch the video or read the blog
Talk soon,
John
P.S. If you need help with any of these hardware decisions, then you can get help from me and other experts inside the Hardware Academy.
