"AI in CAD" gets talked about a lot, but the conversations we have with engineering teams tend to be more specific: what does it actually do differently, where does it genuinely save time, and where is the hype ahead of the reality? Having worked with manufacturers rolling out AI-assisted design tools, here's our take on what's real, what to expect, and what needs sorting out first.
What we mean by AI product design
In a traditional CAD workflow, an engineer manually defines every piece of geometry and every constraint. AI-assisted design tools change that relationship: rather than replacing the engineer's judgement, they take on the exploration and evaluation work that would otherwise eat up hours, generating, checking and ranking options at a scale no individual could manage by hand.
That matters because the complexity engineers are dealing with is outpacing the time available to deal with it. A single component can carry thousands of interacting constraints, structural loads, thermal limits, tolerances, material availability, certification requirements, and reviewing all of that by hand simply doesn't scale. AI tools don't remove the engineer from the loop; they let that judgement operate on a much larger set of options.
Where it actually helps
In-the-moment guidance. Modern AI tools can flag structural, thermal or interference issues while a design is still taking shape, rather than weeks later at a formal design review. That shift, from catching problems after the fact to catching them as they happen, is one of the more underrated benefits.
Generative design and optimisation. Set the performance targets, constraints, materials and manufacturing process, and generative tools will explore thousands of geometric variants that satisfy them, surfacing options an engineer might never have arrived at manually. It doesn't replace design thinking; it widens the pool it draws from.
Faster performance analysis. Embedded simulation lets engineers assess how a design will behave under real load, thermal and fluid conditions far faster than running full traditional FEA on every candidate, so more time goes into the promising options and less into ruling out the weak ones.
Automating the repetitive work. Tolerance analysis, variant configuration, feature look-up, drawing generation: the tasks that consume hours without requiring much judgement are exactly where AI tools tend to deliver the fastest, least controversial wins.
How a generative design workflow actually runs
It's worth being concrete about this, because the workflow is still fundamentally engineer-led. Typically it looks something like:
1. The engineer defines the problem: load capacity, thermal targets, weight limits, manufacturing constraints and any regulatory requirements.
2. The AI explores the design space, generating a large set of geometry candidates that meet those requirements to varying degrees, including options outside conventional design thinking.
3. The engineer narrows the field, weighing manufacturability, cost and assembly considerations the AI can't fully judge, then runs shortlisted designs through higher-fidelity simulation.
4. AI tooling increasingly assists the move from concept geometry to detailed, production-ready models: feature creation, tolerance propagation, drawing annotation.
5. The engineer approves the final design and releases it into the organisation's PLM system for governance and manufacture.
Where the impact is landing hardest
Aerospace and defence: generative design to strip structural weight, with AI simulation speeding up flight-critical certification analysis.
Automotive: applying AI to drivetrain assemblies, motor mounts and thermal management, where weight, safety and regulatory trade-offs are tightly interlinked.
Medical devices: faster iteration on implantable components and surgical instruments, where dimensional precision isn't negotiable and verification timelines are being measurably compressed.
Industrial equipment: optimising frame structures and part geometry to cut material and machining time without giving up structural integrity.
What needs to be in place before you scale it
None of the above happens automatically. In our experience, the organisations that get the most out of AI-assisted design are the ones who deal with a handful of foundational issues early rather than after a stalled rollout.
Data quality. AI tools are only as good as the CAD data they learn from. Inconsistent modelling, missing metadata and legacy files that don't parse cleanly all show up in the output. Data readiness is very often the longest phase of a rollout, not the AI itself.
Integration, not another silo. AI tooling needs to sit inside the CAD environment engineers already use, and connect through to PLM, so AI-generated designs stay inside existing change control and approval processes rather than living in a separate app that needs exporting and re-importing.
Trust and change management. Unfamiliar-looking generative geometry makes plenty of engineers understandably sceptical at first. That trust is built through explainability and hands-on experience, not a mandate from above.
Knowing the limits. Current AI tools are strong within well-defined problems and weaker on genuinely novel challenges or incomplete requirements. Treat it as a capable collaborator, not an oracle.
Governance and data residency. For regulated sectors especially, cloud-based AI tooling needs checking against export control, IP and data residency requirements before it goes anywhere near production data.
Where we can help
If you're evaluating what AI-assisted design would actually look like inside your engineering environment, whether that's a first look at Creo's generative design and simulation tools, or a broader conversation about data readiness and PLM integration, we're happy to talk it through and, where useful, arrange a demonstration.