Design has always been one of the areas where AI struggles to match human ability. Yet most people still misunderstand what it actually means to design with AI. It does not mean handing everything over to the machine. The idea, the direction, the feeling, and the creativity still need to come from you. AI is there to help you execute.
In short, the process looks like this: you come up with the idea. You gather references. You turn your thinking into prompts and use AI to bring it to life. You can use Cursor, Codex, Claude, or any other tool. Essentially, you are translating your thoughts into instructions. AI executes them beautifully, but genuine creativity is still missing. It knows the rules of design — spacing, typography, color theory, hierarchy — but it doesn't understand taste. It doesn't know what looks original, meaningful, and genuinely good.
That's the gap. Real taste can't be handed off to AI yet, but the limitation can be worked around.
Method 1: Use a design skill
This is the fastest route. If you're short on time or the task is small, lean on ready-made design skills — sets of instructions and guardrails that steer AI toward better output. A few well-known examples in this space include Emil Kowalski's Impeccable UI skills and various design-taste skill packages built specifically to curb AI's tendency toward generic output.
The process is simple: connect a design skill, give the AI context, prompt it a few times to produce a site, app, or poster, fix the obvious flaws, and publish. The result is usually decent. It may still resemble other AI-made work, but good skills help you avoid the roughest mistakes and generic "AI slop." Most people use this approach.
Method 2: Design it yourself, together with AI
If you want something that truly stands out, you need to go deeper. You want someone landing on your product to feel that real thought went into it. That requires a more deliberate approach.
Whatever you're building — a mobile app, a desktop app, a website, a landing page, a dashboard, a poster — you need a clear direction first. If you don't know what you're building, AI certainly won't figure it out for you.
Start with meaning
Before anything else, lay the foundation. Ask yourself: Who is this for? What problem does it solve? What should it feel like? What should it express? The interface needs to reflect the meaning of the product. Write it down.
You can use AI here too — but not for design. Let it ask you questions. For example: What is the product about? Who is the audience? What feeling should it convey? What should the brand's character be? What colors and fonts fit? Should it feel minimal, playful, premium, technical, or experimental?
Gather references
Once you have direction, start collecting examples. A strong source is Mobbin, which offers real product screens across many categories. Don't just look at a design and think "this looks good" — ask yourself why. Is it the layout? The spacing? The typography? The structure? The interaction?
Save everything useful and sort it into folders: navigation, first screens, pricing, cards, mobile screens, dashboards, animations, typography. Over time you'll build your own reference library. That's actually how taste is developed.
Other useful sources include Pinterest, Cosmos, Awwwards, Webflow templates, Craftwork, Rebrand Gallery, Component Gallery, Savee, and Lummi.
And one more thing: don't copy a design wholesale. Combine ideas into something that fits your specific product.
Work out the structure
Now define exactly what needs to be built. For a website, that typically means navigation, hero section, features, pricing, testimonials, call to action, and footer. For an app, it typically means onboarding, home, search, profile, settings, core flows, and empty states.
Each block needs clear content. A hero section, for instance, needs a headline, a description, a call to action, and a visual. AI can help with copy, but clarity is still on you.
Build component by component
Now assembly begins. Take individual ideas from your references and adapt them. Instead of saying "build me a whole website," say: "build this hero section in this style, adapted to my brand."
Work step by step: navigation, hero section, cards, buttons, typography, details. AI handles small tasks far better, and you keep control of the outcome.
Generate your own assets
When you need imagery, use image-generation tools. Tools like GPT Images 2.0 or Seedream 5.0 are already strong enough for this. Give clear context: colors, style, composition, and use case. Avoid generic stock photos — custom visuals make everything feel more cohesive.
The workflow
The full sequence looks like this: define the idea, clarify the goal and audience, set the brand direction, gather references, organize them, work out the structure, assemble components, generate assets, add interactions, and polish everything.
The result is far better than anything you'd get from a single prompt.
Method 3: The inspiration board
There's a faster middle ground. Instead of assembling everything piece by piece, give AI a curated set of references. Collect screenshots from Mobbin, Awwwards, Webflow, Craftwork, Rebrand Gallery, Component Gallery, Savee, Cosmos, and Pinterest.
Then tell the AI: "Combine the style and direction of these references into a design for my product. Don't copy directly."
Still give full context upfront, and let the AI ask questions before you generate anything. This is faster than assembling everything by hand, but still much better than a generic prompt.
Three approaches compared
Design skills: fast, decent results. Component-by-component building: slower, but the highest quality. The inspiration board: a balanced middle option.
Extra tools worth knowing
For SVG generation, tools like Quiver make it easier to build animations because elements are separated into layers.
For video, tools like Google Flow strengthen hero sections, backgrounds, and product demos.
For image libraries, platforms like Lummi work well when you need stock imagery — just make sure it matches your brand.
For layering, use background removal and layer stacking to add depth, motion, and interaction. Small details genuinely change a lot.
The bottom line
AI shouldn't replace your thinking. You still decide what to build, why it matters, who it's for, and what it should feel like. AI simply helps you execute faster.
The best results don't come from better prompts — they come from more developed taste.