@MelkeyDev Gpt 5.5 took the coding crown for me in terms of complex codebases or thorough end to end codebase implementation.
Opus still knows how to scaffold simple visual UIs better, but the line stops there. Once GPT catches up on UX polish it's game over for Opus.
@contraben@AnthropicAI The prompt won't save you. The first 1-5 might be fine but eventually it drifts. Regardless of how great the prompt is, and people also get fatigue from prompting and planning and get lazy. That's why I'm building @decantrai . Actually validates against design system drift
I used it for several days. Very intently. It did better on max somewhat. I will say while it did "work". The work it delivered was not the best. In the same context window if I slightly pivoted it would just leave files it already spun up and keep creating new slightly a different versions of the same services. This was not a large context window. I had like 7-10 user services that all did variations of the same thing. I'm sure I could have committed and it would have worked but I'd have like 8 unused service files bloat in my project.
Composer 2.5 even on max destroying my codebases lol. Couldn't even setup basic e2s testing properly. Spun up like 10 unnecessary services randomly without being asked, started messing with unit tests .. not sure why.
Not being mean just being honest I've reverted 95% of the work composer 2.5/Max has done.