Medium: How I'm actually using AI in my UI design workflow in 2026

Rajesh Dey is a UI designer who spent early 2026 testing AI tools in production work. The article documents what worked, what disappointed, and where AI currently hits a wall.

Dey opens with a direct observation: AI design tools, including Figma Make, generate layouts that lack structural integrity. The output needs significant manual reconstruction — fixing responsiveness, spacing, and component alignment. “AI today is more of a rough ideation partner than a production designer,” he writes. The gap between an AI-generated frame and something production-ready remains wide enough to matter.

Despite this, Dey found two applications worth keeping.

Rapid concept generation. AI reaches approximately 70% clarity within minutes. Designers can use that rough draft as a starting point and apply judgment to refine the remaining work. This is fastest in early-stage exploration, where the goal is to generate options quickly rather than finish a polished layout. The AI handles the mechanical first pass; the designer handles the rest.

Design QA automation. AI can audit a file against predefined rules — flagging spacing inconsistencies, detached styles, incorrect token usage — without fatigue. Where a manual review might miss inconsistencies in a complex component library, an AI check runs through the entire file systematically. Dey treats this as a consistency check, not a design judgment.

The article also makes a strategic argument. AI won’t replace designers, but it changes the filtering mechanism. Designers who understand how to direct AI effectively will outperform those who don’t. The core skill becomes knowing which tasks to delegate and which require human judgment — brand nuance, strategic trade-offs, contextual awareness.

Useful for mid-level UI designers who have used AI tools experimentally and want a grounded assessment of where they fit in a professional workflow. Also relevant for design leads evaluating how to introduce AI into team processes without overpromising what the tools can do.