KiS 24/7
KiS Blog

Can AI Design a Real 3D Printed Part?

September 25, 2026 · Watch the 14-minute video or read the short version below

The short version
  • The whole brief was a photo and two numbers. A side-view picture of the bike tube, 90 mm wide and 100 mm deep. That is what the model worked from.
  • The useful part was not the 3D model. It was being told which way to lay each piece on the print bed and which plastic to use, before any filament was wasted.
  • It failed once, on purpose worth watching. The first nut could not physically be tightened. Switching models produced a nut you drive from the inside, and that one worked.

This is part 2 of my Claude and Fusion 360 tests. In part 1, Claude rebuilt two designs from drawings. This time I wanted to know if it could make a part that actually works.

The problem no shop solves

My son wanted to ride with me. The seat was on the bike, the handlebars were on the bike, and there was nowhere to put his feet. Every clamp-on footpeg you can buy is built for a round tube about 26 to 40 mm across. An electric bike hides a battery in its down tube, so the tube is far wider than that. Nothing clamps on.

There was exactly one product that fit, and it was around $300 for what amounts to two footpegs. I had already paid a fraction of that for a seat, pegs and bars together. It was not worth it to me. So the question stopped being which one do I buy and became can I just make this.

What I actually sent

I opened Claude, told it I needed folding footpegs, and said the ones I had did not fit. Then I gave it a photo of the down tube from the side and two measurements: 90 mm across, 100 mm deep. No drawings, no CAD experience, no technical language. I talked to it the way you would talk to a person who was standing next to the bike.

It asked for the sort of thing a machinist would ask for. Whether the sides were flat or rounded. Whether Fusion was open. Then it built the model: a two-piece clamp that wraps the tube, the pegs, the bolts, the pins that hold it all together.

The first design was wrong, and that was fine

The first thing it came back with was a rail system with the pegs built into it. It would have held, but there was no way to fold the pegs up out of the way, which was the whole point. I said so in plain words and it reworked it. A couple of rounds of that and the model was right.

I also told it how much he weighs. It came back with the compression and the sizing and said the pegs would carry him. That is the part people skip when they talk about AI and design: it is not only drawing the shape, it is checking the shape will not snap under a child.

The advice that saved the most time

Here is the part I did not expect to be the most valuable. It told me how to print it.

PartHow to print itWhy
Clamp halvesLay flatThe layers run across the load instead of splitting along it
Bolts and bolt headsStand upThread quality along the shaft
FootpegsLay flat, not uprightMore strength where a foot pushes down
MaterialPETG, not PLAMore give, less brittle under repeated load

Then I sent a photo of my print bed laid out and asked whether I had it right. The answer was mostly no. The clamp halves were correct, the pins were wrong, the pin retainers were wrong, and the bolts were not even on the plate. Ten seconds of checking, against hours of printing something weak.

The bit most people leave out. It did not work first try. Once everything was printed, the nuts would not thread on. The outer edge of the nut was too wide, so it hit the clamp lugs and stopped turning after half a rotation. The model was fine, the fit was fine, and it still would not go together.

How the failure got fixed

I told it the nuts would not spin. The first fix it offered had little wings on the nut so you could grip it by hand, which you cannot actually tighten in that space, because those wings sweep through the same place the lugs are. So I switched to a different model and asked again.

The second answer was the good one: a plain cylinder, threaded for half its length and hollow for the other half, with a hex socket in the open end. It is round, so nothing sticks out to catch the lugs. You drive it with a small Allen key from the inside. It threaded straight on.

The wider lesson there is cheap to apply. If one model keeps circling the same bad idea, hand the same problem to another one. It costs nothing and it is often the whole fix.

What it cost

About 17 hours of printing, spread over a night and the following morning, for both sides. I started on a weeknight and the pegs were on the bike the next day. That is faster than next-day shipping, and the part is one nobody sells. It has been through trails and woods since and it is still on there.

I want to be careful about what that proves. It does not prove AI replaces an engineer. It proves that a person with no CAD background, a printer, and a clear description of a real problem can now end the week with a physical object instead of a saved video.

Why this matters if you run a business

Swap the footpeg for whatever your version of it is: a jig nobody sells, a fixture that keeps breaking, a replacement part with a six-week lead time. The pattern is the same. Describe the real problem in plain words, hand over the measurements you can take with a tape measure, and let the tool do the part you were never trained to do.

That is the same idea behind why AI needs a second brain to be useful. The tool is only worth something when it is pointed at your actual work, with your actual constraints, instead of answering in general.

Not sure where AI would save you time?

The free quiz takes 2 minutes and gives you a starting list for your business. No call, no card.

Take the free 2-minute quiz