AI-enhanced double diamond model
AI-enhanced double diamond model

Prompting the pixel pusher

(how I use AI to ship prototypes, skip blank-canvas anxiety, and feed my god complex)

Why AI at all

  • A starting point. Seeing instead of imagining. That then starts the idea generation.
  • Momentum > perfection. Speeding up every step.
  • Working code early. Real interactions beats a perfect hand-over doc with a janky prototype.

Microsoft’s CPO pushed for similar points in this podcast with Lenny – prototyping with AI allows you to learn a lot faster.

If you aren’t prototyping with AI, you’re doing it wrong. - Aparna Chennapragada, CPO Microsoft

The magic feeling of enablement

I read Peter Koomen's Horseless Carriages and it really resonated with me. Using AI to build is so rewarding while AI squeezed into applications can feel slapped on. When using something like cursor, you're still building it. But instead of moving pixels and drawing rectangles, you're driving it forward, with your ideas and feedback.

Endless opportunities

AI enables so many opportunities, so many doors to open. Design automation, Code generation, User research, Content generation and so on. There's so much optimization and automation that can be done. Saving time, upping quality, impoving the user experience while still increasing the ROI. I've just scratched the surface of how to implement it in my workflow.

Tools I use

ChatGPT, Claude, Gemini, Groq etc.

Most obvious one. I use it for everything from data gathering, exploring ideas, user interviews, everything. Having one model feedback on another model's output is a great way to get a new perspective. Most offer free tiers, so you can try it out before committing.

Midjourney

Generate images with ease, great for moodboards and generating variants of existing images and designs. With stable diffusion getting better and better, it's a great tool to have in your toolbelt (Midjourney). OpenAI's new image generation model, 4o image generation has also turned a few heads and is great for generating images from text, quite a bit slower but a lot more capable when it comes to generating text on images, and small image edits.

v0

Pixel‑pretty and fast, I tend to start here. Uses shadcn/ui and Tailwind CSS as a default, but of course you're free to use what ever framework you want, just be sure to mention it in the prompt and possibly link the docs.

Cursor

With agent mode, and some knowledge about development environments, you are basically all set. Don't know how to do it, ask. Don't know what to ask, ask. Beware of auto-mode and a completely hands-off approach. Although that might work, especially if your using cursorrules, BRDs, PRDs, storypoints, etc Kevin Leneway's approach, Wasp, Playbooks. I find it's best to be hands-on, to learn as well. I also prefer trial and error, it's the best way to learn.

There are a multitude of other tools out there, I haven't tried them all. But some that comes to mind are: Lovable, Bolt, Replit, Tempo Labs, Onlook, Figma make.

All with different approaches to solving the same problem. Allowing you to code with natural language.

Real‑world example: A job tracker

As soon as I have a problem now I tend to go AI first. How can AI help here? When applying for jobs for example. I wanted to keep track of my job applications and the status of each. Sure I could use trello or even sheets, but it was a hassle to update and extract all of the relevant information for each job.

So I started out in v0 to build a LinkedIn job scraper, turned out it was quite hard with Linkedin's anti‑scraping measures. But I started to frame the problem, visually.

  • What would this look like?
  • What's a great user experience when applying for jobs?
  • What are the main issues?

I continued to work on this and was able to build something useful and even got a few people to use it. You can read more about that here.

Kanban board

Considerations

  • Noise. Fast iterations make fast trash – schedule pruning time.
  • Blind spots. AI doesn’t know the edge cases you forgot to mention.
  • Style drift. Being consistent and guiding the AI with rules and docs really helps.
  • Context window For longer sessions, where you go back and forth without improvements, + New chat is your friend.

Where we're heading

The gap between “I have an idea” and “users can click it” is collapsing. Designers who treat AI as a playful co‑author will out‑ship those waiting for perfect specs. Try a weekend build, share the mess, iterate in public. Worst case you learn faster; best case you surprise yourself.

Further rabbit holes