I am a Senior Frontend Engineer, and I want to be honest about something. I still really enjoy building UI by hand. That feeling when a component finally looks right, when a small animation moves exactly how I imagined it — I love this. It did not go away.
But the way I work has changed completely. And the strange thing is that both things are true at the same time: I still love the craft, and my process today looks almost nothing like before.
I want to write about this change, because for me it was not really about the tools. It was about how I think.
The part that did not change
First, let me be clear, because many posts about AI skip this. I did not stop loving to build UI by hand. If you give me a blank file and a nice design, I can happily spend hours there. And this is actually the problem.
When you know your codebase very well — when you know exactly which component to use, which helper already exists, how the spacing works — building it yourself feels like the fastest way. Your hands are faster than your planning. Building manually is not slow because you are bad at it. It is tempting because you are good at it.
This is the trap. And noticing it was my first real change.
What changed: I plan before I build
My process now is almost the opposite of before. Roughly like this:
- First I align on the approach — how we build, how we commit, one shared way of working.
- Then I write a real spec. Not a small note. A proper spec.md with the requirements, the edge cases, and the intent behind the UI, with enough context that someone with no memory of the discussion could still build the right thing.
- Then I break it into small pieces — a features.md with small parts that I can review one by one.
- Then I review and iterate on the plan until it is really good. Most of the thinking happens here, before any UI exists.
- Only after that I delegate the building to the AI.
- And then I keep iterating on the result.
Written like this, it looks obvious. But doing it is not easy. Before, steps 1 to 4 were something I did quickly in my head while I was already typing. Now they are the real work, and the typing at the end is the easy part.
The hard part was not the AI
Everybody talks about prompts and models. For me this was the smallest part. The hard part was in my head.
Going spec-first means I give up the fast feedback loop I love — write code, see it, change it, repeat — and replace it with something slower and more abstract at the beginning: write the plan, question the plan, trust the plan. At the start it felt very uncomfortable.
What helped me was to change how I see my role. Before, I was the author of the UI. Now I am more like a director. I am responsible for the vision, the constraints, the taste, the "no, not like this, like that." But I am not the one placing every single brick anymore. My craft did not disappear. It moved earlier — into the spec, into the review, into deciding what "good" means.
And I did not expect this: it is still creative. Thinking about how a notification should feel when you receive it, what is the best flow for the user who wants to act on it, where there should be friction and where not — this is design thinking. And honestly it uses the part of my brain that I enjoy the most.
Teach the tool, do not babysit it
One lesson came back again and again. When the AI did something wrong in the same way many times — using the wrong tokens, not following the rules of our design system — my first reaction was to correct it every time, turn after turn. But this is just manual work in disguise.
The better way was to teach it once. Instead of fixing the same type of mistake again and again, I spent time to give the tool a real understanding of how our system should be used. After that, the output followed the rules. Same idea as the spec: put your energy into the thing that stays useful, not the thing you will have to redo tomorrow.
Where I am now
If I am honest, the biggest gap in my process is still before all of this: having a real product or design person to help shape the user flow before I open the editor. AI makes the building cheap, so now the quality of the plan is the limit for the quality of the result. A weak plan still gives you a weak result — only faster, and nicer looking.
So the things I keep in mind:
- The hours spent on the spec pay back many times.
- Small, reviewable pieces are what make working with AI possible for me — not one big generated block that I have to understand backwards.
- The instinct to build everything by hand because I am good at it is exactly the instinct I need to notice and resist.
- Time spent teaching the tool is better than time spent correcting it.
I do not think the craft of frontend is dying. I think it is moving. The joy of building a nice UI by hand is still there when I want it. But the leverage, and more and more the interesting part of the job, is now in the plan. Learning to be comfortable with this change, and still keep the part I love — this is what I am working on right now.
And yes, I still go in at the end and adjust the animation by hand. Some habits are worth keeping.
If you want to see more about how I work with AI and build a real leverage in my tech career, I also share things on my YouTube channel: https://www.youtube.com/@anna.vihrogonova.
Come say hi.
/ Anna 💜
Top comments (0)