How to turn your AI into a world-class designer

Ask most people to have AI design something and you get the same purple gradient, text left, image right, cookie-cutter mess. The author spent 12 years leading design teams at Apple, and his take is that AI isn't bad at design, you're just doing it wrong — AI is wired to play it safe, but you can push it out of its comfort zone.

Use random strings for inspiration

Tell an AI to 'make something totally unique' and it gives you the same tired colours and awkward pottery metaphors, because it can't actually be random; it just predicts the most likely next token. The fix is a neat trick: have the AI generate a random alphanumeric string first, then use that string as its design inspiration. Suddenly every run comes back different — new colour schemes, new fonts, genuinely new ideas.

Be bold and fussy as a client

The other lever is to make your prompts specific and wild: 'Make a landing page like a pixel-art game, each section a game screenshot', 'Set it in an isometric 3D city where features are buildings', 'Break every rule with asymmetric layout and dissonant colours'. Stuck for ideas? Ask the AI for a pile of rough seeds, pick the ones you like, refine them with your own taste, then let the AI build it.

Let AIs critique each other

When you iterate, don't let the coding agent review its own work — it will always say 'looks great'. Instead, the author has a stronger model act as a 'design critic': it looks at screenshots only, scores against a clear bar, and the loop isn't done until it independently gives 9/10 or higher. The main model does the work, the critic brings the taste, and the cost stays under 10% of the output tokens.

Then add image and video generation to lift the design beyond cheap gradients, and finally, cut. AI loves to add; good design restrains. Delete the extra glows, the overbuilt buttons, and keep only what earns its place. The author's punchline: AI can surprise you, but you still set the direction and the bar — and don't be afraid to try ideas that sound terrible, because that's often where your underestimated AI really takes off.