Jeff Deehan, a 42 years old Real estate developer in East Lansing, Michigan, described a small steel component designed to secure a camper to his truck, talked it into the GPT-6 Astra’s voice mode, and watched the AI generate a DXF file, navigate to the fabricator SendCutSend, fill out the order form, and pay $156 from saved payment details. Ten days later the part arrived. “In about 10 minutes, I talked a part out of my head into existence without touching my phone,” Deehan told Business Insider.
This sequence is the core of what people call vibe manufacturing: natural language descriptions converted by AI into CAD or manufacturing files, then produced by 3D printers, online manufacturers or small batch manufacturers. The name deliberately echoes Coding’s vibe, the practice of using conversational AI to write software.
Physical errors carry costs that software bugs do not
Unlike code, a flawed physical design produces an object that fails, costs money to reprint, or poses a security risk.
The parallel to Vibe coding is useful but incomplete. Incorrect tolerances, poor dimensions or unsuitable materials cost money to reprint and can create real safety risks. This gap between a quick description and a reliable object is the central tension that ultimately runs through every practitioner in this space.
From robotic arms to MRI scans to children’s toys
A number of manufacturers are already using AI-generated manufacturing files for projects across robotics, jewelry, medical models and consumer electronics.
Amy Zhou and Jake Fitzgerald are developing an open-source text-to-CAD tool for generating manufacturable designs from natural language prompts. According to Business Insider, the project would have raised more than 23,000 GitHub stars at the time of the report.
Fitzgerald used the tool to make a tin robot arm with a reach 0.6 metersmanufactured by SendCutSend for about $200 in parts. Earlier versions produced compound errors, where design errors cascaded over iterations. Fitzgerald reportedly had approx $2,000 on failed parts before the workflow was improved.
Zhou, who previously worked on Apple’s iPhone case team and is now a design engineer at AI rendering company Vizcom, also uses AI to generate jewelry. She continues the work she started at a previous startup that turned uploaded images into 3D wax-cast charms.
Sourish Sharma, a 20 years old in London, uses AI to convert MRI data into a 3D-printable brain model, printing a life-size version and smaller copies for friends. He told Business Insider that the AI workflow requires less manual correction than the open source tool he used previously. This account reflects a single personal project, not any validated clinical standard.
Arthur Prince, an IT worker near Paris, designed a voice-activated children’s toy that generates stories and images in response to a child’s voice. He used GPT-6 Astra for the hardware design and Claude to critique it, then ordered 10 units from the Chinese manufacturer JLCPCB for approx. 500 euros. His daughter tested the first working version.
David Plakon, a tinkerer based near Austin, built a small answering machine with an ESP32 chip, speaker, potentiometer, and 3D-printed housing, inspired by a tin can telephone. He pushed the attachment roughly 15 times while solving fit and assembly problems. The device works, but Plakon does not consider it ready to ship to a factory.
The gap between a prototype and a safe product remains
A working first version and a manufactured, certifiable product are two very different things.
Users remain responsible for checking tolerances, materials, electrical safety, child safety requirements, regulatory compliance and intellectual property. Fitzgerald’s roughly $2,000 in failed iterations illustrates this point financially, and Plakon’s 15 printed attachments do it in plastic. Low-friction design does not automatically mean low-cost or low-risk development.
For simple goods, personalized objects, educational robotics, and early prototypes, Vibe manufacturing can represent a real reduction in the cost of a first experiment. For anything that requires safety certification, manufacturing tolerances, or manufacturing scale-up, professional engineering judgment remains the job, and no voice prompt will change that.