The thing AI changed is not my output quality. It's my ambition.
Before Claude Code and Cursor, I had a mental filter running in the background of every project: is this worth the time it will take? A fluid canvas reveal interaction — interesting, but probably cut it. A fully type-safe component system built from scratch — maybe next time. That filter killed a lot of ideas before they started.
That filter is mostly gone now.
What didn't change is the part people get wrong in both directions.
Taste didn't change. Knowing what you want — what feels right, what's off, what needs another pass — that's still entirely on you. If anything, the stakes are higher. AI executes fast. If your direction is vague, you get a competent version of vague. The output reflects the quality of your input and your ability to review it critically.
Judgment didn't change. I still have to know what good looks like to recognize when I'm not there yet. I still have to catch the thing that's technically correct but subtly wrong — the animation that's 20ms too slow, the spacing that's mathematically consistent but visually awkward. You can't review output you can't evaluate.
What actually changed is the feedback loop and the scope.
Last month I built a canvas reveal interaction for my portfolio — elements animating in from a grid, timed to scroll, with easing that felt considered rather than default. With Tailwind, Framer Motion, and Claude Code iterating in the same session, I had something working in a few hours. Not perfect — I went through probably eight iterations on the easing curve alone. But it shipped.
That used to be a two-day build if it made the cut at all.
The loop being tight matters more than the speed. When I can see it working — or not working — immediately, I iterate differently. I try the contrarian version. I push further. I don't conserve attempts because attempts are no longer expensive.
There's a real distinction between "AI-generated" and "AI-assisted" and it lives entirely in the gap between directing and accepting.
AI-generated is: prompt, ship, move on. The output is the output.
AI-assisted is: prompt, evaluate, redirect, push back, evaluate again. The AI is the hands. The judgment is still yours. That's the work — and it's not easier, it's just different. You're doing less typing and more deciding.
Most people who say AI makes design worse have seen a lot of AI-generated work. Most people who say it's transformative are doing AI-assisted work.
The honest caveat: none of this works if you don't know what good looks like.
AI can't teach you taste. It can help you execute taste faster once you have it. The prerequisite is still years of looking at work better than yours and sitting with exactly why.
That part didn't get automated. It just got more consequential.