A design file can be a useful starting point. The next team still needs to know what was checked, what changed and which assumptions remain open.
AI can help produce a design faster, but a design file is not the same thing as a product ready to manufacture. The handoff between those two is where teams need to pay attention.
On November 3, OpenAI described a prototype device designed with GPT-6 Astra for talking to Codex. The company says two team members worked with the model on the test device as a side project, and a contract manufacturer built a small batch of prototypes. The model produced designs spanning the enclosure, circuit board and software.
The detail worth keeping is what happened next. OpenAI says getting the device ready for manufacturing required several rounds of review and iteration because the model did not consistently account for production limits, tolerances or assembly details. This is one company-reported prototype, not proof that the same process will work for every product or team.
A coherent file is only the start
A model can produce something that looks complete while leaving important constraints unresolved. A part may look right in a rendering, for example, but its fit with another component or its ability to be made still needs examination. Those are different questions from whether a file opens.
An October 17 research preprint assessing GPT-6 Astra on CAD tasks makes a related distinction. Its authors report useful results across several kinds of CAD work, but also describe gaps between plausible geometry and valid, functional, editable models. They observed recurring problems involving geometry, parameter dependencies and relationships between parts. The study evaluates particular test cases; it should not be read as a verdict on every model or design workflow.
Together, the reports suggest a practical boundary: use an AI-generated design as an input to a review process, not as evidence that manufacturing requirements have been met. A prototype can answer some questions, but a small prototype run alone does not establish production readiness.
Make the handoff inspectable
For a team exploring AI assistance in design, the immediate opportunity is to make the output easier for the next person to inspect. Before a design moves to a manufacturer or another engineering group, a handoff can identify:
- The intended function and constraints: what the design is supposed to do, and which requirements the team supplied.
- The files and versions: which model, drawing and supporting documents belong to the same revision.
- Open assumptions: dimensions, materials, dependencies or other details that still need a qualified person to confirm.
- Review evidence: which checks were performed, by whom, and which questions remain open.
- Changes after feedback: what changed in response to review and which related files need to be checked again.
That list is a way to make work visible, not a substitute for a team's engineering, safety or manufacturing procedures. The actual checks depend on the product and the people responsible for it.
Keep the human feedback loop connected
OpenAI also describes work with Chip Ganassi Racing on turning scans of existing parts into editable CAD models and drawings. The company says the racing team's engineers review those outputs for accuracy and manufacturability. That is a different application from the prototype device, but it reinforces the same point: the output has to be useful to the person responsible for the next decision.
A practical first experiment might use a noncritical, representative part and compare the AI-assisted output with the team's existing requirements and review process. Keep the original files and record the corrections needed. Do not send sensitive designs to a tool until the organization has determined that the tool and its data handling are appropriate.
The important question is not whether AI can generate a convincing design. It is whether the team can trace its assumptions, inspect the result and carry it into the next stage with the right evidence. That is what turns an interesting demo into work another person can use.