OpenAI says token usage across its hardware teams has risen 23x since February. Two engineers used Astra to design a Codex companion device in under a week; a manufacturer built a small prototype batch within a month.
But the design still needed several manufacturing reviews. Tolerances and assembly constraints weren't consistently handled.
The test isn't how fast AI generates CAD or PCB layouts. It's whether first-pass manufacturing acceptance improves and total engineering-to-hardware time falls.