How We Use AI at ZÆT

There's a lot of buzz around AI in design these days, but much of it mixes up two distinct concepts: using AI as a tool versus using AI as a designer. At ZÆT, we focus on the former and steer clear of the latter. This difference is more significant than many studios might acknowledge.

Here’s a clear breakdown of how we utilize AI and where its application ends.

The Tools We Actually Use

Vizcom

Vizcom is a tool that converts hand sketches into rendered concepts. When a designer sketches out their ideas; considering proportions, gestures, and intent, Vizcom enhances the visual quality enough to effectively communicate those ideas to clients or evaluate them against alternatives. The creative decisions are made by the designer before using the tool; Vizcom simply speeds up the presentation of those decisions.

Gemini

We employ Gemini for ideation, using it to explore visual directions, generate reference imagery, and test concepts against related ideas. It serves as a thinking tool rather than an output tool, nothing generated by Gemini is presented to clients as a final deliverable. Think of it as a faster version of a mood board session.

Onshape

Our main CAD platform, Onshape, features built-in AI tools that assist with tasks like recognizing features, performing design checks, and handling repetitive geometry. However, it doesn't grasp the reasons behind a feature's shape, the manufacturing processes involved, or the implications of design changes. Those critical engineering judgments remain with the engineer.

Where AI Stops

To be frank, AI stops where the work becomes challenging.

Engineering complex systems requires a deep understanding of how forces move through structures, how tolerances stack up in assemblies, and how mechanisms will perform under real-world conditions. No AI tool we've encountered can handle these intricacies. The geometry it proposes may look plausible, but often lacks physical coherence, a trained engineer can spot these flaws immediately, while someone without that background may not.

Manufacturing constraints present another significant barrier. A design that appears correct in CAD might be impossible to produce at the desired quality, cost, or volume. The reasons behind this are specific, contextual, and often not apparent until you've spent time in factories. Questions like which tooling methods are suitable for a particular geometry, what surface finishes are achievable with specific materials, and where weld seams will land, all of this knowledge is rooted in hands-on experience, not in a dataset. AI lacks this understanding because manufacturing expertise comes from real-world practice.

This isn't a limitation that can be fixed with a better AI model. It's a fundamental disconnect between AI's ability to recognize patterns in existing data and the nuanced judgments required in manufacturing engineering.

Why This Matters for Our Clients

As an industrial design and engineering studio, our clients approach us because they need tangible solutions, not just visual representations. The value we provide lies not in speeding up the initial phases but in the quality of decisions made throughout the entire process, decisions that occur at the factory, during certification, and at the tooling stage.

AI does help us accelerate the early stages. It allows us to explore a wider range of ideas in less time, clarify presentations, and move quickly through the initial definition phase. This is genuinely beneficial.

However, the success of a product rarely hinges on those early stages. It depends on the engineering decisions made in the middle and the execution choices made at the end. Those are human decisions, informed by manufacturing knowledge and accountability. AI doesn't belong in that space, and it shouldn't.

We leverage AI to enhance the efficiency of skilled professionals, but we don't use it in ways that would undermine the expertise that ensures valuable outcomes.