Essay
Building by Discarding
How I iterate with AI
I built 40 versions of one portfolio in 3 days using AI. Most of them failed. Here's what I learned about the process — the dead ends, the pivots, and the frameworks that emerged.
- Explore the medium before you have a message — build things you can feel, not just describe
- How people interact with your design matters as much as how it looks
- When content breaks a design, the fix is usually structural, not cosmetic
- Small changes carry more weight than new features — refinement is its own creativity
- The last breakthrough is rarely more polish — it's a new idea that reframes the whole problem
01 — Discovery
Testing Four Visual Directions Before Writing a Word
Before touching any portfolio content, I built four completely different visual effects. The goal was to start with the canvas, not the words — learn what the medium can do before deciding what to say with it.
Each one created a different mood. The fluid (top-left) felt alive and responsive. The terrain (top-right) felt slow and ancient. The particles (bottom-left) felt cinematic but hands-off. The organic growth (bottom-right) felt alien. These are feelings, not features — and knowing them early meant I could pick the right one later.
With the fluid chosen, the next question: how should a visitor engage with it? I tested three ideas. One (E19) made scrolling change the behavior of the fluid itself — scroll down and the colors warm, the movement gets more turbulent. Another turned the cursor into a magnet that bends nearby elements. A third let you click to shatter and reassemble the layout.
02 — The Pivot
Adding Real Content (and Watching It Break)
This is where most design-with-AI projects fall apart. The visuals work. The interaction feels right. Then you drop in actual text and everything fights. E20 was the first version with real portfolio content layered onto the fluid — and the truth was: content breaks everything.
Then came the review. I scored every variant on visual impact, content clarity, interaction quality, and performance. The verdict was blunt: beautiful noise is still noise. The content was present but not structured. A visitor could feel the craft but couldn't extract the signal.
03 — Convergence
Small Changes That Made the Biggest Difference
The instinct here was to subtract, then reframe. I built four versions of the same design, changing only how intense the fluid was and how the content was laid out. E31 dialed down the motion to feel calmer and more meditative. E36 added directional transitions — content sweeping in from the side instead of fading. Same underlying effect, completely different personalities.
E40 added the final layer: frosted-glass panels, bouncy animations where elements overshoot then settle into place, and something I called Ghost Canvas — a system that records every visitor's cursor movements and replays them later. So when you visit, the fluid you see is being shaped by someone who was there hours ago. The portfolio is never empty.
Takeaways
Five Frameworks
- Start with the medium, not the message
- Scroll is a metaphor, not a mechanic
- Content enters on the effect's terms
- Subtract until it breaks, then add one thing back
- The final 10% is a new idea, not more polish
The process itself was a product exercise — scope broadly, test interaction models, integrate the real requirement, refine through subtraction, then ship something that reframes the problem. 40 explorations in 3 days. Not because speed matters — because momentum does.
View the live portfolio →