I used AI for a week - and this is what I’ve learned

I took a week off to learn Hhow to design with AI

I had been quietly worried that I was falling behind.

Everyone seemed to be talking about vibe coding, AI tools, and using code to create design work. I wanted to catch up, but work was already overflowing. Every day was packed with urgent projects and tight deadlines, leaving little room to learn something new.

I considered experimenting at work, but when you're constantly putting out fires, learning feels like a luxury. Even if a tool might make you faster later, the upfront investment can feel impossible.

I was also burned out. So I decided to take a week off—not to drink margaritas on a beach, but to do something different. I wanted to recover by experimenting, learning, and seeing what these tools could actually do.

The pre-week

Because I wanted to build something personal and unrelated to work, I first needed to decide what to build. That meant figuring out the problem, the user, and the shape of the product—even if the problem was deeply personal.

So whenever I was at the playground with my kid or commuting, I talked with Chatty to explore ideas.

The list came down to a few concepts:

• An AI social media manager for small businesses
• An app for keeping notes about friends so you can show up more meaningfully
• An app that lets you look up history and interesting facts about anything around you as you walk
• An app that helps moms journal and build deeper relationships with their kids
• And many more half-formed ideas

I started with the map/history app because I wanted to experiment with maps and understand how they worked. I developed the problem statement, user story, and early product direction. When my attention drifted, I also explored the other ideas.

At the same time, I looked through Pinterest and Behance for visual inspiration, created mood boards, and started thinking about the brand direction for each app.

Day 1: A clunky start

I had my PRD. I had Claude Code set up. I felt ready.

Then reality hit: I had no idea how to build an iOS app.

I asked Claude what to do, and it started walking me through the setup. I installed Xcode and Simulator, deleted apps to make room on my computer, and slowly worked through technical issues. Eventually, I got a map to load.

But the interactions I assumed would come for free—panning, zooming, basic map behavior—were not as straightforward as I expected. Using a map in Simulator felt different from using one on my phone, and every fix took a long time. Map-related issues were much slower for Claude to work through than text or visual changes.

Most of the day disappeared into troubleshooting maps, Simulator issues, and performance problems. My computer was not powerful enough to support the process smoothly, so I kept looking for hacks just to keep things moving.

By the end of the day, I had basic map interaction working, but I was nowhere near designing the actual app.

Day 2: Starting over

The next morning, I had more clarity.

My goal was not to become an iOS engineer. My goal was to learn how to design with Claude. I wanted to create screens, flows, and prototypes—not spend the entire week fighting Xcode.

So I paused the map project and switched to the small-business social media idea.

I spent the day iterating the PRD I started and, in the process, learned a surprising amount about social media marketing and San Francisco bakeries. I was already comfortable using AI for writing, but I was less sure whether it could help me create a reasonable UI flow.

Based on the PRD, I worked with Chatty to define the main pages, subpages, and user flow. Then I asked it to generate an HTML version of the product. It was not perfect, but it made the idea feel real.

When I opened Figma, I realized I still did not know what the app should feel like. I needed a brand guide.

So I went back to Chatty and worked through the brand direction. From there, I asked AI to help create a color system, which I developed into a color token system. I did the same for typography, spacing, and sizing.

Once those foundations were in place, I finally felt like I had something solid to build from.

Day 3: Building momentum

With the design system in place, building screens became much easier.

Using atomic components, I moved back and forth between Figma and Claude Code and built the full screen flow in one day.

Claude Code definitely made me faster. But did it make me 10x faster? Not exactly.

Part of that was because I was not letting AI fully take the wheel. I still reviewed every decision: layouts, icons, hierarchy, and visual choices. AI helped me move faster, but I was still the designer.

Day 4: Critiquing with AI

One of the best parts of the week was realizing that AI could act like a design partner.

Because Claude had my project context saved in claude.md, I could ask it to critique a screen or flow with the full product in mind. That was especially useful when I could feel something was off but could not yet name what needed to change.

Claude was also surprisingly good at finding missing tokens and retrofitting them back into the design system. It spotted inconsistencies across components and asked for confirmation before making changes.

Accessibility was another area where AI was genuinely helpful. It flagged color contrast issues, missing supporting text, and places where the interface could be clearer.

Day 5: Exploring visuals

After a few days, the main flows were solid enough that I wanted to explore visuals and interaction design.

I started experimenting with mascot ideas, animation concepts, and more expressive directions. That was when I realized Claude and ChatGPT were not always the best tools for visual-heavy generative work, so I started looking more seriously at Midjourney and other image-focused tools.

One day was not enough to fully explore that side of AI, but it was enough to see the potential. It made me want to keep experimenting.

Reflections

Before this week, I felt overwhelmed by AI. I was scared I was falling behind.

After spending a week with these tools, I felt the opposite. I became more convinced that AI can be especially empowering for designers. I could prototype faster, explore more directions, and push further into interaction and visual design than before.

It also made me see product design as an even more strategic role.

AI made the production work faster, but it did not make me a 10x designer. The details still needed taste. The product direction still needed judgment. The sharpest decisions still had to come from me.

What changed was the shape of the process.

Instead of designing one screen at a time, I could critique the whole product at once. Instead of moving slowly from idea to wireframe to prototype, those steps started to collapse into each other. The process became more fluid, iterative, and holistic.

The takeaway is simple: I do not need to be scared of AI. I need to make it part of my process.

Not as a replacement for design judgment, but as a way to expand what I can explore, test, and build.