OpenAI is testing whether AI can help engineering teams turn design ideas into working prototypes faster. For businesses, the potential savings depend on how much human review and correction those designs still require.
The company on Nov. 3 outlined engineering experiments involving GPT-6 Astra and other OpenAI models, covering hardware design, electronics, simulations and vehicle development. The projects include an AI-designed device, expanded work with Chip Ganassi Racing and a new research partnership with Mazda.
One experiment involved building a device that allows users to communicate with Codex through a dedicated interface.
Two OpenAI team members used GPT-6 Astra to develop the prototype in less than a week as a side project. Astra handled the device’s appearance, mechanical enclosure, internal structures, electrical components and printed circuit board (PCB) design. It also generated the user interface, firmware and software integrations needed to connect the device to Codex. A contract manufacturer produced a small batch of prototypes in under a month.
However, preparing the design for manufacturing required several rounds of review. Astra struggled to account for production limitations, component tolerances and assembly requirements. OpenAI says these shortcomings will help improve future models.
OpenAI reports growing AI use across its hardware teams
OpenAI’s hardware and infrastructure teams already use Codex and ChatGPT to create computer-aided design (CAD) models, prepare manufacturing drawings, configure simulations and review designs.
OpenAI says token consumption across its hardware teams has grown 23-fold since February. That measures usage rather than productivity or cost savings. OpenAI also reported results from BenchCAD 2.0, which evaluates models on reconstructing 3D components and assemblies and identifying circuit board connections. A newer internal model achieved an average score of 0.875, compared with 0.594 for GPT-6 Astra at its highest reasoning setting.
The results suggest progress in engineering capabilities, although benchmark performance does not guarantee production-ready designs.
Mazda and Chip Ganassi Racing join the experiments
OpenAI says it is expanding its collaboration with Chip Ganassi Racing (CGR) to explore converting 3D scans of existing components into editable CAD models and manufacturing drawings. This could help engineers recreate parts when original design files are unavailable. CGR engineers will review the results for accuracy and manufacturability.
A separate research collaboration with Mazda focuses on vehicle aerodynamics and crash safety. The teams are exploring airflow and structural simulations to evaluate vehicle designs while considering styling and development costs.
What businesses should test before adopting AI engineering tools
For engineering and IT leaders, the practical starting point is a limited pilot with a clearly defined task, such as preparing a drawing for review. Compare the AI-assisted process with the existing workflow, using the same inputs and acceptance criteria.
Measure the time required to produce an approved result, including corrections and engineering review. Track model spending, integration costs and rework alongside speed. Producing an initial design faster offers limited value if employees spend the saved time repairing it.
Check whether the tools work reliably with the team’s CAD software, file formats and version-control processes. Review what proprietary design information the model can access, how that information is handled and who can authorize changes to engineering files.
Keep qualified engineers responsible for validating dimensions, materials, tolerances and manufacturing requirements. Any pilot involving safety-related components should follow the organization’s established testing and approval procedures before designs move into production.
As with other enterprise agentic AI deployments, integration and oversight belong in the business case. Expand the pilot when it demonstrates measurable savings after review and correction while meeting existing quality standards.
Want to use AI more effectively in your daily workflows? TechRepublic readers get free 7-day access to The Neuron Academy, our practical learning platform for AI tips, prompting techniques, and workplace skills. Browse all lessons →
Let us teach you How to Talk to AI for free! Try our six-minute course at The Neuron Academy and learn a few simple ways to write better prompts and get more useful results from AI, or browse our other AI course for free for seven days. Check out all the lessons here →