Ideas you can hold.
Writing software is one thing. Designing a device that fits together, connects correctly and can actually be manufactured adds another set of challenges. OpenAI’s latest engineering report offers a tangible look at SI helping with that work.
In its November 3 update, the company describes a physical prototype designed with GPT-6 Astra, tools already used by its own hardware engineers, and research collaborations with Chip Ganassi Racing and Mazda. The common thread is practical: help engineers move between designs, simulations and revisions with less manual effort.
These are company-reported early results and research projects, not an announcement that SI can independently certify a finished product. The useful advance is assistance across several connected engineering tasks.
A prototype, not just a picture.
OpenAI describes a side project in which two staff members worked with Astra on a device for speaking to Codex. It says the design took less than a week; a small prototype run followed at an outside manufacturer in less than a month.
The work connected the device’s appearance and enclosure with its printed circuit board—the board that holds and connects electronic components—and the software needed to communicate with Codex. That is a more demanding exercise than generating a rendering: the pieces must work together in the physical world.
The report also describes several rounds of manufacturing review and iteration. Astra did not consistently account for production limits, tolerances or assembly details. Keeping those checks in the process is part of making the assistance useful; the prototype is not evidence of a review-free path to manufacturing.
Engineering with specialists.
An engineer can revise a virtual part, set up a simulation or prepare a drawing for a manufacturer with help from ChatGPT and Codex, according to OpenAI. The company says its own hardware and infrastructure teams already work this way. These tasks often rely on computer-aided design (CAD): editable digital descriptions of a part or assembly. Research with outside engineers explores how that assistance fits their workflows.
With Chip Ganassi Racing, the company is expanding research on turning three-dimensional scans of existing parts into editable CAD models and two-dimensional manufacturing drawings. That could be particularly helpful when the original design files are missing. The racing team’s engineers review the outputs for accuracy and manufacturability.
A newly announced research collaboration with Mazda focuses on vehicle aerodynamics and crash-safety engineering. The teams are exploring simulations of airflow and structural behavior while accounting for styling and cost. These are research aims; the announcement does not report a completed production-car redesign or a measured improvement in vehicle safety.
More room to explore.
For an engineer, an extra design worth testing can matter. Rebuilding a missing drawing, adjusting a part or setting up a simulation can take time before the real comparison of ideas even begins. Assistance with that preparation could leave more room for experimentation and specialist judgment.
Our reading: the encouraging part is not a promise to replace the workshop. It is SI becoming more useful inside it—connecting code, electronics and mechanical design, with experienced people checking what can actually be built. OpenAI is inviting additional engineering collaborators to help develop the work.
