Over the past year, I've spent a significant amount of time on one question:
figuring out exactly what AI can help me do—and what it can't.

Not at the level of "I tried Midjourney and got some nice-looking images."
I mean genuinely embedding AI into my daily design workflow,
hitting its limits over and over, and recalibrating each time.

This article is my current-stage summary—
not a tutorial, but a set of working principles that are actually effective for me.

I. My Fundamental Stance on AI Tools

AI is not a tool that replaces designers.
It's a lever that amplifies a designer's judgment.

I've said this many times, but it needs unpacking.

"Amplify" means: if your judgment is strong, AI helps you land decisions faster and more precisely. If your judgment is weak, AI just helps you make bad decisions faster. So AI doesn't lower the barrier to entry—it accelerates divergence. Good designers get stronger with AI. Weak designers get exposed more thoroughly.

II. My AI-Assisted Workflow

I break my workflow into four phases. AI plays a different role in each:

1. Information Deconstruction

When I pick up a new project, I start by using AI to "break down" the information.
I feed it requirement documents, competitor screenshots, and user feedback,
and ask it to surface the core tensions, potential design blind spots, and edge cases that might be overlooked.

From practice: While working on the UCOMP autonomous driving operations platform, I fed a 20-page requirements document to Claude and asked it to list every scenario involving "abnormal states." It organized 17. Three of them were edge cases I had completely missed.

2. Solution Divergence

This is where AI is most capable—and where the most caution is required.
AI can generate 10 layout options, 20 interaction logics, and 5 information architectures in seconds.
But the quality of those outputs depends entirely on the quality of your prompt.

My principle: use AI to diverge, use your own judgment to converge.
I never take AI's first suggestion and run with it directly.
But I use its output as "cognitive friction material"—
I look at its logic, then ask myself why I disagree. That process is often more efficient than staring at a blank screen.

3. Content Production

UI copy, error messages, empty-state text, onboarding prompts—
these used to take me a lot of time to craft.
Now I have AI generate five versions first, then choose the closest one and refine it manually.

This phase saves me roughly 60% of the time.
It also made me realize: my judgment in this area is actually stronger than I thought—
I can quickly identify which version "feels right." What I was missing before was just the raw material to choose from.

4. Validation and Iteration

Once a design is ready, I feed the screenshots and design rationale back to AI
and ask it to play the role of a "skeptical user" or a "doubting PM" challenging my decisions.
It doesn't replace real user testing—but it helps me patch obvious logical gaps before I get to testing.

III. What AI Cannot Do

After all this use, I'm also very clear about where AI's limits are.

It cannot sense the emotional weight of context.
When designing an interface for emergency system takeover, AI can give you a sensible layout—
but it cannot understand the pressure an operator feels at 3am, alone, watching a system fail.
That perception only comes from humans.

It cannot make value judgments.
When business goals conflict with user experience, AI will hand you two sets of solutions—
but it won't tell you which matters more. That judgment always belongs to the designer.

It cannot build trust.
Ultimately, users trust the brand behind the product.
That trust is built through consistent, human design decisions accumulated over time.
AI can help you execute, but it cannot build that accumulation for you.

IV. My Conclusion

I'm not worried about AI replacing designers.
But I am worried that designers who refuse to use AI
will be replaced by designers who know how to use it well.

That's not alarmism. It's already happening.

AI is a mirror.
It reflects your judgment, your thinking boundaries,
and the design blind spots you haven't yet noticed.
Learning to collaborate with it is essentially forcing yourself to know yourself more clearly.

And that, I believe, is the most important work a designer can do in the age of AI.