How to Use AI for Design: 4 Lessons We’ve Learned

Column

From bad prompts to good drafts: 4 takeaways from building AI into our design workflow — without losing taste or control.

We didn’t add AI to our design process to chase hype. We did it because staring at a blank screen wastes time.

Since then, it’s helped us unblock ideas, explore more directions, and ship faster. But it came with a learning curve.

Here are 4 things we’ve picked up along the way, courtesy of Jeza.

AI won’t replace designers — but it’s a great creative sidekick

AI hasn’t replaced anyone on our team. It’s just helped us explore faster.

We use it to generate variations, find inspiration, and get unstuck. Think of it like a teammate who’s always down to brainstorm — and never runs out of ideas.

Instead of redrawing the same layout five times, we generate ten options and cherry-pick the interesting ones. The human still makes the call. But now we get there faster.

The first draft is usually mid — keep going

We learned this early. The first result is often average. The tenth? Much better.

This is where most people get stuck. They try it once, don’t love the output, and give up. But AI is an iteration engine. You’re not outsourcing quality. You’re outsourcing the grunt work.

Treat it like compounding. The more you prompt, refine, and direct it, the closer you get to something good.

Good prompts are clear, not clever

“Make something cool” is not a prompt. It’s a shrug.

We had to learn to be specific. The more we added constraints — tone, color, format, reference brands — the better the output got.

Here’s a quick example:

Prompt	Result
“Design a modern header image”	Generic gradient, stock layout
“Minimalist hero section, dark background, no photos, for a SaaS startup like Linear.app”	Much closer

Prompting is a skill. It takes reps, but it’s worth learning.