The positioning shift here is interesting. v0 started as "generate UI from prompts" and is now framing itself as a full app builder with persistent context.

What I find compelling about this direction: the bottleneck in AI-assisted development is not the initial generation, it's the iteration loop. Having a tool that maintains context across changes and understands your codebase holistically is genuinely more useful than one-shot generation.

The real test will be how it handles the messy middle - when you're 70% done and need to refactor, integrate with external APIs, or handle edge cases that weren't in the original prompt. That's where most AI coding tools fall apart in my experience.

Curious how they're handling the economics here. AI-first tools tend to have brutal unit economics at scale - inference costs add up fast when users expect instant iteration. The shift to Claude and "optimized model selection" suggests they're thinking about this carefully.

That last point where you demo your product while fulfilling requirements from customer indeed sounds promising while very scary in reality.