Should welders worry about AI? A guy who does both weighs in.

I run a metal fabrication shop in Chicago. Twenty years on the floor — gates, fences, staircases, architectural steel. I also build AI systems, and I mean actually build them: the quoting app my own shop runs on, I wrote. So when someone asks me whether welders should worry about AI, I’m not guessing from either side of the fence. I live on both sides of it.

Short answer: worry is the wrong move. So is ignoring it. Here’s the honest breakdown.

Nobody’s coming for your bead

Let’s kill the fear first, because most of what you’ve read about AI and the trades was written by people who’ve never held a grinder.

AI is software. It reads, it writes, it does math fast. It does not fit up a frame that came back from powder coat a sixteenth out. It doesn’t shim a gate post because the concrete guy poured proud. It doesn’t look at a hundred-year-old Chicago brick wall and know the anchors are going to find rubble. It doesn’t tack, doesn’t feel heat warp coming, doesn’t run an overhead bead in a stairwell where the drawing and the building disagree.

Yes, robots weld. They’ve been welding in factories for decades — long, straight, repeatable production runs, thousands of the same part. That’s manufacturing automation, and if that was going to take your job it would have taken it already. Custom fabrication is the opposite of repeatable. Every job I quote is a one-off in a building that’s never square, for a customer who changes their mind twice.

The part of this trade that pays custom money — the judgment, the hands, the fix on the fly — is exactly the part AI can’t touch. As the desk jobs get automated, we’re going to be the last ones standing, physically building things like we always have.

What AI genuinely changes: the office half of your shop

Now the part you shouldn’t ignore. Because there IS a half of your business AI is very good at, and it’s the half you probably hate: quoting, paperwork, scheduling.

I know because I built one. I fed my quoting app a half-finished fence description — on purpose, to see if it was BS — and it flagged the real mistake: a cantilever gate needs two heavy carriage posts plus a catch post, not standard fence posts. That’s a whole post, and real money I’d have left on the table. Then it built the materials list, the cut list, the build sequence, landed the labor — about 16 hours of site install at $145 an hour, roughly a $15K job — and generated two PDFs with one click: a branded proposal for the customer and a shop sheet with the cut list for the crew.

That used to be my evening. Now it’s minutes, and I check its work instead of typing it.

The same goes deeper than speed. Tell it what’s actually in your shop and it quotes around your reality: a Millermatic 220 has a duty cycle and a range of what it can and can’t weld. An automated bandsaw that runs up to 18 cuts at a time prices a cut list differently than one guy setting a miter for six cuts, banging those out, then the straight cuts. Time is money, and your tools change the number.

And here’s the part that should actually get your attention — the money you’re leaving on the table right now. Around Chicago I watch guys charge $60 an hour while the other shops are at $100, $125, $145 — plus consumables, plus materials, plus the four hours somebody burns pulling permits that never lands on an invoice. That’s not the customer’s fault. That’s a quoting problem. The real value of AI in a shop isn’t speed; it’s that it stops forgetting the costs you keep forgetting. Price like you mean it.

Where the AI still gets it wrong

I’d rather you trust this than be wowed by it, so here’s the other side — and I’m describing my own product.

It won’t catch every question there is. Sometimes it won’t ask about permit fees. Sometimes the number is a ballpark and you need to know that it’s a ballpark. When it misses, I go add that data to the job description myself. It’s a tool you steer, not a machine pretending to be perfect — and anyone selling you the perfect machine is selling you something.

What a welder should actually do about it

Not a motivational list. A job list.

1. Keep getting better with your hands. The weld is the moat. Nothing in this post changes that — it’s the reason the rest of this list is worth doing.
2. Point AI at your quoting first. It’s the biggest bottleneck and the fastest payback. Every hour of paperwork it takes is an hour back on the floor — or back home.
3. Charge the real number. Rate, consumables, materials, permit hours. If a tool helps you find the costs you’ve been eating, it just paid for itself.
4. Run the BS-detector on every tool. Take one job you already know the price of and run it through. If it can’t handle your real work in an afternoon, drop it and move on. No subscription is worth babysitting.

The welders who should worry aren’t the ones reading about AI. They’re the ones still quoting at $60 an hour on the back of an envelope while the shop across town figured out the office half. Don’t be that shop.