ChatGPT and Codex Accelerate Physical-World Engineering Development

Engineering R&D is undergoing an efficiency revolution driven by generative AI. Recent practical cases disclosed by OpenAI show that ChatGPT and Codex are being used to accelerate physical-world engineering workflows such as CAD modeling, PCB design, and simulation analysis, covering multiple stages from conceptual design to validation and deployment. The significance of this direction lies in the fact that AI application scenarios are extending from pure software into hardware engineering, where they must contend with real materials and real constraints.

From Code to Drawings: Codex Fills the Gap in Engineering Automation

For a long time, engineering automation has relied more on specialized scripts and parametric tools, which have high barriers to entry and are hard to reuse. Codex's value lies in translating natural language intent into executable code and scripts, letting engineers drive modeling and design processes with descriptive language.

In CAD scenarios, engineers can use Codex to generate parametric modeling scripts, batch-process geometric changes, and write automated check logic, thereby reducing repetitive operations. PCB design benefits similarly: routing rule validation, bill-of-materials organization, and batch processing of design files can all be semi-automated through AI-generated scripts. These tasks share a common trait—clear rules and high repetitiveness—which is exactly what code generation models excel at.

Parametric Modeling is a core paradigm in the CAD field: design objects are not fixed geometric shapes but adjustable parameter sets composed of dimensions, constraints, and feature history; modifying parameters automatically rebuilds the model. It enables batch geometric changes, design series, and automated checks, and is also the prerequisite for Codex-generated scripts to act directly on CAD workflows—what the AI outputs is essentially code that manipulates parameters and constraints, not stroke-by-stroke drawing of shapes. Only by understanding this can one see clearly where the interface lies between code generation models and engineering modeling tools.

The Engineering Knowledge Collaboration Role Played by ChatGPT

If Codex is responsible for "hands-on" work, ChatGPT takes on more of a "thinking" collaboration role. Engineering problems often span materials, processes, electrical, and structural domains, and engineers need to repeatedly search and compare across large volumes of specification documents, datasheets, and historical designs.

In such scenarios, ChatGPT can serve as a knowledge interface: explaining design specifications, sorting out the causes behind simulation results, assisting in drafting technical documentation, and translating cross-domain terminology into language the team can understand. Simulation workflows are especially typical—pre-processing setup, boundary condition selection, and result interpretation all require experiential judgment, and AI can shorten the learning curve for newcomers in these areas while also helping senior engineers quickly validate ideas.

Simulation workflows are typically divided into three stages: pre-processing, solving, and post-processing. Pre-processing includes geometry cleanup, mesh generation, and setting material properties and boundary conditions; solving is performed by the solver; post-processing involves interpreting and judging results such as stress, temperature, and flow fields. Among these, pre-processing setup and result interpretation are highly experience-dependent and are also the main sources of error. The value of AI in these stages is not to replace the solver but to make implicit experiential rules explicit, helping engineers complete setup trade-offs faster and make preliminary attributions for anomalous results.

Why AI Adoption Is Harder in Physical-World Engineering

Compared with pure software tasks, physical-world engineering imposes stricter constraints on AI. A design error is not just an error message; it may mean prototyping costs, safety risks, and project delays. This means AI output must be reviewable and traceable, and engineers must still be accountable for the final result.

Judging from the direction disclosed by OpenAI, its strategy is not to let AI replace engineers in decision-making but to embed AI into existing workflows to take on auxiliary tasks such as script generation, document processing, and knowledge retrieval. This "human-in-the-loop" model better fits the risk preferences of the engineering industry, and it also explains why CAD, PCB, and simulation have become priority scenarios—they all have mature toolchains and clear validation methods.

"Human-in-the-loop" means AI output must be reviewed and confirmed by humans before proceeding to the next step, with humans retaining final decision-making authority and responsibility. In engineering, this model is not only a matter of risk preference but also involves compliance and traceability requirements: design changes need approval records, and errors need to be traceable to specific stages. Therefore, AI exists more as an advisor and draft generator, and its output is incorporated into existing review and validation processes rather than directly driving manufacturing or construction.

The Practical Impact on Engineering Teams

In the short term, the most direct change is that repetitive work is compressed, allowing engineers to devote time to trade-off analysis and innovative design. In the medium term, the way engineering teams organize knowledge may change: design specifications, script templates, and simulation experience will be沉淀 (accumulated) in AI-callable forms, becoming team-level knowledge assets.

For individual engineers, the ability to clearly describe requirements and accurately judge the quality of AI output will become a new dividing line in capability. Tools lower the execution barrier, but the demand for engineering judgment has not decreased—rather, it has become more prominent.

Routing rule validation in PCB design is usually performed by Design Rule Check (DRC), which automatically scans the layout based on constraints such as line width, spacing, vias, and impedance and reports violations. Traditionally, rules must be configured manually one by one, but AI-generated scripts can batch-organize constraints, convert rule formats, and preliminarily classify check results. Similarly, organizing the bill of materials (BOM) involves cross-checking model numbers, packages, and alternative parts—a task with clear rules but tedious—making it a typical scenario where code generation and knowledge retrieval work together.

Conclusion

The entry of ChatGPT and Codex into CAD, PCB, and simulation workflows marks that generative AI is moving from digital content production toward physical engineering collaboration. This path will not be achieved overnight, but its direction is already clear: AI does not replace engineers but frees engineering teams from tedious execution, returning professional judgment to the core position.

One-line Summary

ChatGPT and Codex are being embedded in a human-in-the-loop manner into CAD, PCB, and simulation workflows, accelerating physical-world engineering development.

Paragraph Summary

The article introduces the practical direction disclosed by OpenAI: ChatGPT and Codex are being used to accelerate physical-world engineering workflows such as CAD modeling, PCB design, and simulation analysis. Codex focuses on translating natural language intent into parametric modeling scripts, batch processing, and validation logic, taking on execution work with clear rules and high repetitiveness; ChatGPT serves as a knowledge interface, assisting with specification explanation, simulation result interpretation, and technical documentation drafting. The article points out that errors in physical-world engineering are costly, and AI is embedded in existing toolchains in a human-in-the-loop manner rather than replacing engineer decision-making. In the short term, repetitive labor is compressed; in the medium term, team knowledge will be accumulated in AI-callable forms, and engineers' ability to describe requirements and judge output will become a new dividing line in capability.