I've been a UX engineer for 12 years. I spent the first decade of my career watching the tools around me change — from Photoshop to Sketch to Figma. Each transition had true believers and skeptics. Each time, the tool changed what was possible but didn't change what made great design great.
AI tools are doing the same thing, faster.
Here's my honest take after 12 months of integrating them into my daily workflow at Insphere AI.
Cursor AI: The One That Actually Changed How I Work
I was skeptical. I'd tried GitHub Copilot, found it useful for boilerplate but actively misleading for anything architectural. Cursor is different — specifically because of how it handles context.
The "Apply" feature understands your codebase. When I'm building a new component at Insphere AI, I can describe the component in natural language and reference existing components for style consistency — and the suggestion actually works in context, not just in isolation.
What it changed for me: I write frontend code 40% faster. I spend less time on autocomplete bikeshedding and more time on the actual UX problem. The keystrokes saved let me focus on what matters: is this interaction right?
What it didn't change: I still catch structural issues, accessibility problems, and performance antipatterns that Cursor misses. AI writes code; engineers decide what code to write.
Claude for UX Writing and Research Synthesis
The UX writing workflow is where I use Claude most. I conduct user interviews, synthesise notes into themes, and draft microcopy — and Claude handles the first passes that used to eat 20% of my research time.
Specifically useful:
- Affinity mapping from interview notes: Feed raw interview transcripts, ask for theme clustering. The output is a solid first draft — not final, but it reduces the blank-page problem.
- Microcopy variants: "Write 5 versions of this empty state message for a prayer tracking app designed for Muslim users in the UK." The constraints narrow the output quickly.
- Usability report drafting: Structure of method, findings, recommendations — AI can draft the scaffold, I write the insights.
Figma AI: Useful in Two Specific Scenarios
Figma AI's autocompletion is useful exactly where Figma itself used to be slowest: layout suggestions and component naming at scale.
Where it helps me: renaming 40-layer components consistently. Building quick layout variations from a starting frame. Not for ideation or design decisions.
Where it doesn't help: anything requiring taste. Figma AI doesn't know your product's users. It doesn't know your brand system. It generates mediocre generic UI that you'd redesign anyway.
The Honest Summary
AI tools in 2025 are excellent productivity multipliers for well-defined tasks and still weak at tasks requiring judgment, taste, and stakeholder context.
If you're using them to write code faster — they work. If you're using them to think through design problems — you're outsourcing the part that makes you valuable.
The best UX engineers I've worked with use AI as a first-pass generator and immediately apply critical judgment to the output. That's the workflow that actually saves time without degrading quality.
