Vibe manufacturing’: how hobbyists are prototyping real hardware with AI
Vibe coding gave non-developers a way to ship software without writing it. Now the same pattern is showing up in physical goods. A loosely connected group of hobbyists, engineers, and indie builders is using AI to design parts, print cases, and order fabricated components — and they’re calling it vibe manufacturing.
The workflow is straightforward: describe what you want to a chatbot, iterate on the generated design file, then send it to a fabricator or a home 3D printer. No CAD training required, or so the pitch goes.
Here are five real projects that show where this stands today, including what it cost, what broke, and what actually shipped.
️ A truck part, designed and ordered in 10 minutes
Jeff Deehan, a 42-year-old real estate developer in East Lansing, needed a small steel piece to secure a camper to his truck. He described the part and its mounting angle to GPT-6 Astra in voice mode. Five minutes later, the chatbot had a prototype design.
Deehan told GPT-6 Astra about SendCutSend, a custom sheet-metal fabricator. The chatbot exported a DXF file, navigated to SendCutSend’s site, filled in the order form, and completed the $156 purchase using stored payment details. The part arrived 10 days later and fit correctly.
“In about 10 minutes, I spoke a part out of my mind into existence, without touching my phone.” — Jeff Deehan
Deehan expected at least a millimeter of error. There wasn’t one. He acknowledges this was a simple part and that results vary with complexity.
A text-to-CAD tool with 23,000 GitHub stars
Amy Zhou and Jake Fitzgerald are building an open-source text-to-CAD tool that already has over 23,000 stars on GitHub. Zhou is a design engineer at AI rendering startup Vizcom and previously worked on the iPhone cases team at Apple. Fitzgerald pivoted from crypto back to robotics. They collaborate between San Francisco and New York.
Zhou uses the tool to make jewelry. Her previous startup converted photos into 3D charm models, which were then cast in wax. It made a necklace for Cluely CEO Roy Lee before she left the company. She still makes AI-rendered charms for friends who want jewelry featuring their pets.
Fitzgerald’s project is more mechanical. He used the text-to-CAD tool to design a robot arm in sheet metal and paid roughly $200 to have it fabricated by SendCutSend. The arm has 0.6 meters of reach, includes a motor and a camera, and is built as an educational product for learning robotics fundamentals.
The early iterations were expensive. Fitzgerald described a period of “compounding failures” where bad AI designs stacked on top of each other — loose screw holes were one example. He spent about $2,000 working through bad iterations. He reports that GPT-6 Astra resolved those issues.
An MRI scan turned into a 3D-printed brain
Sourish Sharma, a 20-year-old from London, received MRI data from a research study three years ago and decided to print it. He used an open-source tool to convert the data file into a printable format, but the results were rough. “Some of the folds of the brain were not very foldy,” he said.
With GPT-6 Astra, Sharma prompted a conversion of the same data. The result was, by his assessment, more anatomically accurate than the open-source tool produced. He’s printed a life-size version and several miniatures he gives to friends.
“AI literally does replace CAD engineering completely. It’s kind of unfortunate if you’re a CAD engineer. Fortunately, I’m not.” — Sourish Sharma
Kids’ toys: the sweet spot for complexity
The vibe manufacturers are converging on one category where the complexity stays manageable: children’s products. Simple enclosures, single-function electronics, offline operation. No cell service, no browser, no compliance nightmare.
Arthur Prince, a 34-year-old IT worker outside Paris, built a toy that generates stories and images from a child’s voice prompt. He developed everything from the casing to the internal wiring using GPT-6 Astra, then asked Claude to critique the design. Once both chatbots reached a consensus, he ordered 10 units from Chinese manufacturer JLCPCB for roughly 500 euros, unsure whether they would work.
They did. His daughter was the first tester. He describes her reaction as pride that her father made something like that.
David Plakon, a 39-year-old tinkerer in Austin who has owned a 3D printer since 2015, had a different constraint: he wanted his kids to leave voice messages for grandparents in Florida without putting them on the internet. He and his wife Lara built an answering machine styled after a tin can phone.
Plakon gave GPT-6 Astra design constraints: all parts had to be either 3D printable or purchasable off the shelf. The chatbot produced a parts list including a potentiometer, a speaker, and an ESP32 chip, then walked him through assembly. He’s printed the case roughly 15 times, iterating each time. Early versions had pieces that didn’t fit together.
“AI’s good but not perfect.” — David Plakon
The box works. Plakon plans to get it into his kids’ grandparents’ home for Christmas. He is not under the illusion that it’s ready for factory production. “I don’t have any delusion that I could send this to a factory, and they would be able to make it,” he said. “For now, it’s very fun to prototype.”
What the pattern actually looks like
Across these five projects, a few things hold:
- Simple geometry ships faster and cheaper. Deehan’s truck bracket worked first try. Fitzgerald’s robot arm cost $2,000 in failed iterations before the design stabilized.
- Multiple AI passes help. Prince ran GPT-6 Astra and Claude in sequence, using one to check the other’s work.
- Budget for reprints. Plakon printed his case 15 times. Fitzgerald spent $2,000 on bad fabrication runs. Factor iteration into your cost estimate.
- Offline, single-function products are the current ceiling. Nobody is vibe manufacturing a car. Kids’ toys, custom brackets, robot arms, and jewelry are the realistic scope.
- The open-source tooling is moving fast. Zhou and Fitzgerald’s text-to-CAD tool crossed 23,000 GitHub stars, which signals real developer adoption, not just hype.
The CAD engineering question is real. Sharma’s assessment is blunt. Zhou and Fitzgerald are building tooling that explicitly targets that skill gap. Whether vibe manufacturing stays a hobbyist pattern or scales into something that changes how small-batch physical products get made will depend on how well the design accuracy holds as product complexity increases.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
