Software people have long joked that hardware is hard. OpenAI just tested that joke by having its newest model design a physical device from scratch, and the result is a decent prototype plus a reminder that atoms are still less forgiving than pixels.
From prompt to PCB in under a week
According to The Decoder's account of an OpenAI blog post published November 3, the company's hardware team used GPT-6 Astra to design a device for talking to Codex, OpenAI's coding agent, as a side project. OpenAI says Astra handled every step: the enclosure, the circuit board layout (including picking components and routing traces), the firmware and the user interface. The design took less than a week, and a contract manufacturer built a small batch of prototypes in under a month.
It wasn't a one-shot miracle. The project needed several rounds of revisions because Astra "didn't always account for manufacturing constraints, tolerances, and assembly details." OpenAI also says its hardware teams' token usage for CAD models, manufacturing drawings and simulations is up 23 times since February. On BenchCAD 2.0, a benchmark that tests whether models can reconstruct 3D parts and identify connections on circuit boards, GPT-6 Astra scored 0.594 at its highest reasoning setting, while an unreleased internal model scored 0.875. The Decoder notes OpenAI hasn't shared any details about that internal model.
The hard part was always the last 10%
A week from idea to manufacturable design used to need a small team and a lot of coffee, so the speed is real progress. Look closely at where Astra stumbled, though. Tolerances, assembly and manufacturing constraints are the parts of engineering that live in experience, supplier quirks and the tribal knowledge of whoever has watched a run of enclosures warp. The model got the big picture right quickly and still needed people and a factory to sort out the physical details.
That split will sound familiar to anyone who has used AI to write code. You get the first draft fast, and you still need someone who knows where production breaks. The value moves from typing out the design to reviewing it, which makes experienced reviewers more important, and OpenAI's own revision rounds show why.
AI can draft the gadget now, but the people who know why the screw holes won't line up still have plenty of work.
If you're working out where AI can speed up your own builds and where you still want an experienced engineer reviewing the output, tell us about the project and we'll help you draw that line.
Source: The Decoder