The experiment started as a provocation aimed at myself. I had spent months talking about AI and design, writing about tools, comparing workflows. But I was still designing the same way I always had. Figma, component libraries, stock photos. The usual. Then I asked myself an uncomfortable question: if I truly believe AI changes design, why am I not proving it?
That is how the 30-day challenge was born. The rules were simple: every design deliverable I produced during one month had to involve artificial intelligence in a meaningful way. Not as decoration or novelty. As part of the real working process.
Week 1: productive frustration
The first days were terrible. Not because AI did not work, but because I did not know what to ask it. I was used to thinking in pixels and components. AI thinks in concepts and directions. I had to change the way I frame problems.
My first real project was a landing page for a fintech. Normally I would start searching references on Dribbble, building a mood board by hand, selecting typefaces. This time I asked Claude to help me structure the value proposition before touching Figma. The result was revealing. Instead of designing first and justifying later, I was justifying first and designing with intention.
For images I used Midjourney. And here came the first shock: generating an image takes 30 seconds. Generating the right image can take two hours. Execution speed does not equal decision speed.
Week 2: patterns start to emerge
By the second week I had a rhythm. I discovered that my most efficient flow was: think with Claude, explore visually with Midjourney, execute in Figma, iterate with ChatGPT for microcopy and text variations.
That week's project was an investment dashboard. I used Claude to map user flows and prioritize which information to show first. The conversation led me to a hierarchy I would not have found alone: the most important data point was not portfolio performance but the change since last visit. An insight that came from asking AI about retail investor psychology.
With Midjourney I generated the onboarding illustrations. I discovered that flat isometric styles produce more consistent results than character illustrations. Generated characters always have something that feels uncanny. Geometric shapes do not.
Week 3: the turning point
The third week was where everything changed. I had three simultaneous projects and for the first time I felt that AI gave me a real speed advantage. Not because it did the work for me, but because it eliminated the parts of the process where I was not adding unique value.
Searching stock photography: eliminated. Midjourney generates exactly what I need. Writing the first draft of microcopy: eliminated. ChatGPT produces ten variations in seconds. Researching UX best practices for a specific pattern: eliminated. Claude gives me the state of the art with sources and reasoning.
What remained was what truly matters: making decisions. Choosing between options. Evaluating whether something works for the specific user. Adjusting emotional tone. Defending a creative direction to the client.
Week 4: the reflection
On the last day of the month I took inventory. I had delivered six complete projects. In a normal month I deliver four. Quality had not dropped. In some cases it had risen because I had more time to iterate on important decisions instead of spending it on mechanical tasks.
But the biggest learning was not about productivity. It was about identity. For thirty days I had to constantly redefine which part of the work was mine and which part belonged to the machine. And the answer surprised me: my part was more important than I thought. AI can generate everything. But it cannot decide what to generate or why.
As I wrote in my first post, design did not die. It evolved. After this month I can say that is not a motivational phrase. It is a literal description of what I experienced.
The designer who comes out of this experiment is not faster. They are different. They think in systems, direct tools, and focus on what machines cannot do: understand people.