Meet the people who “make vibes” and bring their ideas to life

Jeff Deehan is a real estate developer by day and a hobby guy by night.

The 42-year-old has a workshop in his East Lansing home. One project remained elusive: a small piece of steel to attach a camper to his truck. As Deehan played with OpenAI’s GPT-6 Astra in voice mode, he described the part he needed and the angle at which it would be placed. Five minutes later, the chatbot had made a prototype.

Deehan told GPT-6 he knows of a company called SendCutSend that makes custom sheet metal. The chatbot downloaded its design as a DXF file, opened the SendCutSend website, filled out the form, and paid the $156 with the payment information it had stored. Ten days later, he was at her door.

“In about 10 minutes, I brought a part of my mind into existence, without touching my phone,” he said.

Deehan is part of a growing class of enthusiasts who are building physical gadgets with the help of AI. They use chatbots to create prototypes, then send the AI-generated files to manufacturers, shop makers, and their home 3D printers.

Modeling physical objects with AI carries risks. Deehan was sure his prototype would be off by a millimeter here or there, although it ultimately turned out just right. Other ambiance makers said their products required multiple iterations, eating into their budgets.

Standards have long used AI to write software, now dubbed “vibrational coding.” These manufacturers say AI is also coming to hardware creation – and they call it “ambience manufacturing.”

A Growing CAD Disaster

Amy Zhou and Jake Fitzgerald are building an open source tool for creating ambiances. They call it “text-to-CAD” – and it already exceeds 23,000 stars on GitHub.

Zhou previously worked on the iPhone case team at Apple and is now a design engineer at Vizcom, an AI rendering startup. Fitzgerald wanted to get back into robotics after moving into cryptography. They text new ideas between San Francisco and New York.

The duo hopes to disrupt CAD or computer-aided design. CAD engineers use software to design and optimize 3D models. This requires training and skills. Ambiance makers say you can do this easily with a great language model.

Zhou makes jewelry with AI. The 25-year-old startup turned images uploaded by users into 3D models of charms, then cast the models in wax. The company made a necklace for Cluely CEO Roy Lee before Zhou left the company. She still occasionally makes AI-rendered charms for her friends when they want necklaces featuring their pets.

Fitzgerald is more interested in making robotic arms. Using the text-to-CAD tool, the 31-year-old designed the arm out of sheet metal and paid about $200 to have it built by SendCutSend. It has a range of 0.6 meters, is equipped with a motor and a camera, and he considers it an “edutainment” product to learn about robotics before the rise of the industry.

A few months ago, Fitzgerald saw “cumulative failures” when the AI ​​built bad designs on top of bad designs. The screw holes have come loose, for example. He spent about $2,000 ordering parts multiple times. With GPT-6 Astra, he said this is no longer a problem.

Sourish Sharma saw AI changing CAD. The 20-year-old from London had an MRI three years ago for a research study and requested its data. He got a 3D printer and used an open source tool to turn the data file into something printable. The tool didn’t work very well, so he had to modify it by hand. “Certain folds of the brain were not very pliable,” he said.

Now Sharma has a much bigger printer – and AI. With just one prompt, GPT-6 Astra blew the open source tool out of the water, converting the data and creating what it considers a more accurate brain. He printed a life-size one, as well as a handful of miniatures that he gave to his friends.

“AI is literally completely replacing CAD engineering,” Sharma said. “It’s a bit unfortunate if you’re a CAD engineer. Fortunately, that’s not the case.”

The first audience: children

The more complex a product is, the more difficult it is to manufacture. You won’t be able to create a car at home with ChatGPT anytime soon.

There is a population whose products are often simple and accessible: children. Arthur Prince, a 34-year-old computer scientist from the Paris region, said the children’s toy he modeled in a week with AI would have taken him months with CAD knowledge alone.

Prince’s toy creates stories and images with AI, based on a child’s voice message. He develops everything from the case to the internal wiring with the GPT-6 Astra, then asks Claude to critique its operation. Once the two chatbots reached a consensus, he ordered 10 toys from Chinese manufacturer JLCPCB for around 500 euros, fearing he had wasted his money on something that wouldn’t work.

Fortunately, they did. Prince’s first tester, his daughter, loves it. “She’s very proud to see that I can achieve something like this,” he said.

The advantage of Vibe manufacturing children’s toys is their simplicity. Many parents don’t want their children to surf the web. So AI prototypers don’t have to worry about those pesky questions about cellular services and web browsers.

David Plakon wanted his children to leave messages for their grandparents in Florida, but he didn’t want them on the Internet.

The 39-year-old from Austin is a longtime “handyman.” He’s had a 3D printer since 2015 and was printing his kid a “KPop Demon Hunters” sword as we spoke. So he and his wife, Lara, thought they could build their own answering machine in the style of the Tin Can telephone.

Plakon gave the GPT-6 Astra design inspiration and instructions for building a prototype. All the parts had to be able to be purchased or built with his 3D printer, he told the chatbot. He easily assembled a model, had him purchase a potentiometer, speaker, and ESP32 chip, and guided him through the assembly.

He has now printed the design around fifteen times, improving it each time. Initially, some pieces didn’t fit together. “AI is good but not perfect,” he said.

But the box works and he hopes to sell it one day. For the moment it remains on a small scale, with the aim of getting the answering machine from the grandparents for Christmas.

“I don’t believe I could send this to a factory, and they would be able to make it,” Plakon said. “For now, it’s a lot of fun making prototypes.”