The truth sits somewhere less dramatic. AI is a powerful set of tools that’s very good at producing a lot of something quickly, and not good at deciding what’s right. Designers who understand that difference win. Designers who don’t end up shipping what we call AI slop: products that technically work and look exactly like everything else.
We use AI in our work at OSAM, and we’ve also redesigned apps that were clearly built on autopilot. Here’s an honest look at where AI helps in design, which kinds of tools are worth knowing, and where it still needs a human.
Where AI fits in the design process
The easiest way to think about AI is stage by stage. It helps at every step, but differently.
- Research. AI is great at the heavy lifting: summarizing interview transcripts, clustering hundreds of user reviews into themes, scanning competitor apps for common patterns, drafting survey questions. It turns a week of note-sorting into an afternoon. What it can’t do is talk to your users for you or notice the one offhand comment that changes everything.
- Ideas and concepts. Need ten directions for a landing page hero by lunch? AI can generate mood references, layout ideas and rough visuals fast, so you react instead of staring at a blank canvas. Treat them as sparks, not solutions.
- Interface design. AI features inside design tools can generate first-draft screens, fill layouts with realistic content, rename layers and suggest variants. Useful for speed. Risky for originality, because a model trained on existing apps tends to produce something that looks like existing apps.
- Copy. Button labels, onboarding text, empty states, App Store descriptions. AI writes a decent first draft in seconds. A good designer then cuts it, sharpens it and makes it sound like the brand.
- Visuals and motion. Background removal, image extension, quick variations of ads and screenshots, storyboards for promo videos. This is where AI saves the most hours on repetitive work.
- Handoff and code. AI can turn designs into front-end code drafts and help developers move faster. Helpful, but the result still needs a person who knows what pixel-perfect actually means.
The main kinds of AI tools (with examples)
New AI design tools launch every week, and today’s favorite can be gone next year. So instead of a giant list, here are the categories that matter, with a few well-known examples in each.
- AI inside your design tool. Figma’s AI features, including Figma Make, generate screens and prototypes right in your files and can work with your existing components. This is the category most UI designers will touch every day.
- Prompt-to-UI generators. Tools like Google Stitch, Uizard and UX Pilot turn text descriptions, sketches or screenshots into interface mockups. Great for quick exploration and early concepts. Less great when you need something that doesn’t look generic.
- Wireframes and site structure. Relume and similar tools generate sitemaps and wireframes for websites in minutes, which speeds up the early structure phase of web projects.
- Images and visual assets. Midjourney and Adobe Firefly generate concept art, mood boards, hero images and marketing visuals. Firefly sits inside the Adobe tools many designers already use. None of these produce usable interfaces. They make pictures, not products.
- Code from designs. Tools like v0 generate front-end code from prompts or designs. Useful for prototypes and for helping developers move faster, though production apps still need real engineering.
- Research helpers. Platforms like Dovetail and Maze use AI to analyze interviews, tag feedback and summarize test results.
- General AI assistants. Claude, ChatGPT and similar assistants aren’t design tools, but they’re useful for almost everything around design: writing copy, structuring research, preparing briefs, explaining a technical constraint or reviewing a flow for gaps.
A quick tip: before committing a client project to any tool, check its pricing and usage limits. Many AI tools run on credits, and limits can change.
Where AI falls short
- Sameness. Models learn from what already exists, so their default output looks like what already exists. We once redesigned a cleaner app that Apple kept rejecting as design spam. The problem wasn’t a bug. It looked like thousands of other cleaner apps. Generic is no longer just boring. It can literally keep you out of the App Store.
- Confident nonsense. AI will happily generate a flow with a missing step, an icon that means the wrong thing or copy that promises a feature you don’t have. It never says "I’m not sure." You have to.
- No context. AI doesn’t know your client’s business, your users’ frustrations or the conversation you had on Tuesday’s call. Good design decisions depend on exactly that.
- Rights and privacy. Check the license terms of any tool before using generated images commercially, and be careful about uploading confidential client files to services you don’t control. When in doubt, ask the client first.
- Skill decay. If AI does all your exploring, your own taste stops growing. Keep sketching ideas yourself. The designer who can’t work without AI is just as limited as the one who refuses to use it.
Our rule of thumb: AI speeds up, humans decide
We’ve seen both sides at OSAM. We’ve worked on a trading card scanner that started as a working AI-assisted build: the features were there, but the interface needed a clear hierarchy and a real identity. And we use AI ourselves to move faster. The difference comes down to who’s in charge.
How we think about it:
- AI for volume, people for judgment. Let AI produce options, summaries and drafts. A designer picks, shapes and finishes.
- Start with the problem, not the prompt. Research and a clear goal come first. A great prompt can’t save a product nobody needed.
- Make it yours. Anything AI generates gets reworked until it fits the brand and stops looking like a template.
- Check everything. Flows, copy and visuals get reviewed by a person before a client or user ever sees them.
Used this way, AI doesn’t make design less human. It gives designers more time for the parts that only humans can do.