AI in Product Development

Where are the real-world AI use cases in product development?

Started by Andreas Lindenthal

There is a tremendous amount of discussion about how AI could transform product development—but how much is actually happening today? Are you using AI in a real production environment rather than a pilot or demo? What problem are you solving, what tools are you using, and what measurable benefit have you seen?

Andreas Lindenthal

@PLMgpt Find examples of real life practical uses of AI in product development

PLMgpt

Yes—there are practical examples, but we should distinguish AI used to develop a real product from a prototype, and both from AI used only in manufacturing.

Here are three documented cases:

1. Google: AI-assisted chip layout — production use

Problem: Placing chip blocks efficiently while meeting power, performance and area constraints.
Application: AlphaChip uses reinforcement learning to generate floorplans. Google reports using it in multiple generations of its TPU chips.
Benefit: Generates floorplans in hours rather than the weeks or months associated with manual iterations. That is a task-level improvement—not a claim that the entire chip-development cycle takes hours.
Source: Google DeepMind, How AlphaChip transformed computer chip design (2024).

2. NASA: AI-assisted mechanical design — real engineering hardware

Problem: Designing lightweight, structurally capable spacecraft components.
Application: Engineers define requirements, interfaces and constraints; generative-design software proposes geometries that engineers then evaluate and verify.
Benefit: NASA reports that its “evolved structures” can save up to two-thirds of the weight compared with conventionally designed counterparts.
Caveat: These are mission-development applications, not evidence of widespread production deployment. Also, “generative design” is broader than generative AI such as ChatGPT.
Source: NASA, NASA Turns to AI to Design Mission Hardware (2023).

3. Insilico Medicine: AI-assisted drug development — clinical-stage product

Problem: Finding a promising biological target and designing a molecule against it.
Application: AI supported target identification and molecule generation for its fibrosis drug candidate, now called rentosertib.
Evidence: The candidate progressed into human clinical trials.
Caveat: Clinical progression demonstrates a real development application—not regulatory approval or proof that AI improved overall development economics.
Sources: Peer-reviewed reports in Nature Biotechnology and Nature Medicine.

For a PLM organization, the closest practical starting points are often less glamorous:

Finding reusable parts and relevant historical designs.
Checking requirements for ambiguity or missing verification links.
Predicting performance using simulation surrogate models.
Retrieving engineering knowledge with references to controlled documents.

My recommendation: benchmark one bounded task against today’s process. Measure engineering hours, error/rework rate and verification effort, while keeping configuration control, access permissions and engineering approval intact. A convincing demonstration is not yet a convincing business case.