Most "AI in design" content stays at the surface — screenshots of a chat prompt generating a UI. Aufait UX's writeup goes further into the parts that matter for anyone shipping product: how AI output integrates with the actual handoff to engineering.
Output format matters more than the demo Figma Make outputs editable frames, not flattened screenshots — meaning design output stays inside the working file rather than needing to be redrawn from an image. It also generates component code and style references, which speeds up dev handoff considerably compared to tools that only produce static comps.
Model selection isn't a footnote The team uses Gemini 2.5 Pro for complex reasoning-heavy flows, Flash models for fast iteration cycles, and Claude Sonnet when they need a balance of structural logic and creative output quality. Worth remembering that "AI design tool" isn't one undifferentiated category — the underlying model choice changes what you get.
Mokkup.ai exports only as JPEG/PNG — no conversational refinement, no editable component output. Good for early wireframe-level thinking on interactive ui design for data-heavy dashboards specifically, not for anything downstream in the pipeline.
The case study that stands out: a health-tech R&D platform needing a custom report-generation flow, where the team had already studied Salesforce and Power BI for reference but kept producing designs that either borrowed too heavily from existing patterns or didn't actually reduce interaction load. A prompt synthesizing their research plus design thinking got Figma Make to a structural thread they hadn't considered — which then became the basis for the actual build.
That's a meaningfully different use case than "AI generates a landing page," and it's the kind of example worth referencing when your team is evaluating whether a given AI tool earns its place in a real ux design services pipeline or just adds another revision loop nobody asked for.
The piece also covers a finding that matters for how teams scope AI use generally: a community platform project where field research overturned assumptions an AI personalization model would have shipped on by default — specifically around device ownership patterns in smaller Indian cities. No model flagged it. A researcher in the field did.
For dev and design teams collaborating around ui ux design services, this kind of tool-by-tool, output-format-aware breakdown is more useful than the usual "AI will change everything" framing — it actually tells you where in the pipeline each tool belongs and where it doesn't. Full breakdown AI in UI/UX Design — Hype vs. Reality