Chapter 1 of 3
8th May 2026
The app went live today. Here's what the final stretch looked like — Flutter warnings, AI tradeoffs, and why I design the UI last, not first. User Experience is the only focus.

Tap the picture to see it full size
Two Opus 4.7 sessions open, two projects in motion — that's been my rhythm lately. Most of the day belonged to the regular work, the kind that pays the bills. But Hurroz, my personal venture, phase 1 is already finishing up. Yesterday it was almost there. All that remained were the small things — minor UI edits, a few Flutter warnings nagging from the corners, an unused variable here, a context error there. Nothing dramatic. It all came together quietly, and as I write this, the app is making its way up to the Play Store. Fingers crossed.
The new design is striking, maybe even a little bold for a journaling app. Whenever I sit down to generate code with Opus 4.7, it wants to talk first — questions, then more questions, and somewhere along the way the whole afternoon is gone, spent explaining the vision. The AI has grown slower in its execution, even as the results have grown sharper. I keep thinking — enough with the questions. So I ran a small experiment: I stopped pushing back and simply accepted whatever defaults it recommended. To my surprise, it worked. The feature came out perfectly. But here's the catch — Claude can write code, but it cannot design. The UI it produces is cluttered, dense, and clearing it up takes ages.
Gemini is the opposite. It has a real eye for design, but carrying code from one chat into another is always a coin toss with it. So I've settled into a workflow that suits me: build the feature first, get it working, then hand the visual direction over later. People sometimes ask why I don't design the UI before writing the code. Honestly, I believe an architecture-led UX serves the end user better. Once the bones of a feature are working, that's when the real design begins — moving things around, finding the rhythm, letting the screen breathe.
Supabase has made all of this gentler. Compared to projects where I had to build out a whole Python backend from scratch, this feels almost like a relief. I find I enjoy the front-end work more anyway. Though I'll admit — Streamlit in Python still has a way of turning out the most beautiful tables.
By the time I finished writing this journal, the app went live.
Written by
Sulabh RastogiMore by Sulabh Rastogi
Short link: hurroz.com/c/AR1U1
Yes sir...I agree...the gemini model gives beautiful code and UI experience but the underlying structure is reliable on claude but the issue is that claude has a lot of limit issue(for me)