I started "development" with sonnet 3.7 and 4.0 which was terrible and slow. Used cursor 20$ and claude 20$ plans together with warp 20$ plan. Every time a model was free in cursor (grok, early gpt, anonymous) i used them in parallel for days or weeks, for maximum efficiency. One day free gpt 5.0 introduced tanstack query because he thought it would be useful... :D Honestly the app looked terrible, and design changed a lot on the way, especially I continuously kept up coming with new ideas which is a terrible practice! :D Eventually Sonned 4.5 was a huge quality jump, so I subbed for 100$ claude, and was able to work 8-10 hours a day with it. Then they came out with opus 4.5 included in that plan, which was ... kind of exponentially better in all terms. So I worked even more, started to make bigger improvements and features on one shot, and It all worked. At the end I was helped by senior dev who reshaped infrastructure and some security stuff, but most of the app is created by agent, only me supervising them. Curious what opus 5 will bring, and how the app finds its audience!
At least... I can use my dream test management app finally! :)