Like someone mentioned, the loops are a nightmare. Honestly, Claude Code has been pretty disappointing for full-stack stuff lately. I switched over to Cursor AI Code Editor Pro for better context handling, and it kills the constant re-writes. Also been testing Windsurf Code Editor AI lately, which feels way more reliable for complex app architecture. TL;DR: save your credits and stick to a proper AI IDE instead of raw CLI tools.

Honestly, I tried going all in on Claude Code yesterday and it was a bit of a headache. Everyone says focus on architecture, but ngl, it just gets stuck in loops when things get complex. It kept rewriting the same component until I burned through way too many credits. Had issues with state management specifically, which was a huge pain. Here is how I see the current landscape compared to what people hype up:

• Anthropic Claude 3.5 Sonnet
• Incredible at drafting logic, but unfortunately it tends to hallucinate API routes if the file structure gets too deep. Use it for one-off helpers, not the whole app.
• Cursor AI IDE
• Honestly feels way more stable for full-stack work because it actually understands the context of your whole codebase. It doesnt lose the plot like standalone Claude usually does.
• Vercel v0
• Great for getting the UI skeleton down quickly, but dont trust it with your backend logic. It breaks things constantly if you push it too hard. If you want my two cents, stop trying to make the AI the architect. It is not as good as expected at holding the big picture. Just write the boilerplate yourself or use something like Cursor AI IDE to manage the heavy lifting. It keeps the context local and you wont end up with broken code at 2am. It is just way more reliable that way, imo.