Designing with AI Without Losing Your Judgment

Every AI design tool promises the same thing: paste a prompt, get a UI. The result looks fine at first glance. Then you try to use it.

That's the problem. Designing with AI is easy. Designing well with AI requires something the tool doesn't have: judgment. And judgment isn't a plugin you install — it's what separates a design that works from one that only looks good in a screenshot.

I've spent months integrating AI into my design process. Not as a replacement, but as an accelerator. What I've learned is that the question isn't "can AI design?" but "when do I let it design and when do I step in?"

The "good enough" trap

AI-generated designs have a subtle problem: they look polished. Correct border radius, consistent spacing, harmonious color palette. Everything looks professional. But when you look closely, something is missing.

Intentionality is missing.

AI follows trends, not user needs. It generates the same gradient hero you've seen on 200 landing pages. It applies uniform spacing without considering which elements need room to breathe and which need to be grouped. It creates generic visual hierarchies instead of hierarchies that guide the user toward the right action.

It's not that the output is bad. It's that it's indifferent. It doesn't take a stand. It has no opinion about what matters most on your screen. And in design, that's a serious problem.

When to accept AI suggestions

There are contexts where speed matters more than originality:

• Data-heavy layouts — Tables, dashboards, list views. The structure is predictable and the AI nails it.
• Repetitive patterns — CRUD forms, settings pages, configuration flows. Problems solved a thousand times where reinventing adds no value.
• Responsive variations — Adapting a desktop design to tablet and mobile. AI handles reflow rules well.
• Exploration prototypes — When you need 5 quick variants to evaluate directions, not to ship.

In these cases, AI eliminates mechanical work. Accept the output, adjust the details, and move on. Your time is worth more elsewhere.

When to override

There are moments where "good enough" isn't enough:

• Brand-defining screens — Your main landing, your onboarding, your first impression. Here every pixel communicates who you are.
• Onboarding flows — The user's first experience defines whether they come back. AI doesn't understand the emotional tension of a first encounter.
• Emotional connection moments — Purchase confirmations, error messages, empty states. Where the human tone makes the difference.
• Product innovation — If you're creating something new, AI can only recombine what already exists. It can't imagine what hasn't been done yet.

In these contexts, use AI to generate a starting point and then rebuild with intention.

My framework: Generate, Evaluate, Decide

After months of iteration, my process boils down to three steps:

1. Generate multiple options fast. Don't marry the first one. Ask for 3-4 variants. AI is fast — use it to explore the possibility space, not to reach the final answer.

2. Evaluate against your design principles, not against each other. Comparing variants against each other leads you to pick "the least bad." Comparing them against your principles (clarity, consistency, accessibility) leads you to pick the right one.

3. Decide with intention, not laziness. "It's fine" isn't a design decision. "This works because the hierarchy directs attention to the CTA and the spacing groups related elements" — that's a decision.

The difference between a designer who uses AI well and one who doesn't is in step 3. Both generate options. Only one decides with judgment.

Practical filters: 5 questions before accepting

Before shipping any AI output to production, I run it through five filters. If it fails more than two, it goes back to the drawing board.

These questions aren't theoretical. I use them literally on every review. I have them on a sticky note next to my screen. They sound simple, but most AI-generated designs fail at least one.

The most revealing question is the third: "Would you defend this in a review?" If the answer is "it's what the AI generated," that's not a defense. If the answer is "it works because it solves X problem in this way," you can ship it.

The designer's new role

AI is the fastest junior designer you'll ever have. Prolific, tireless, with good surface-level taste. But it needs your direction. Without it, it produces designs that look good and perform average.

Your job has shifted. It's no longer spending 3 hours tweaking padding and picking colors. It's deciding what problem to solve, how to solve it, and evaluating whether the solution actually works. You went from execution to curation.

And that's not less valuable — it's more. Because execution can be automated. Judgment can't.

---

This is the sixth article in the series. The first was about the tools I use. The second about why they fail and how to fix it. The third about how to build a skill system. The fourth about the Figma-to-code workflow. The fifth about MCP servers and tool connections.