Yes, the 10x productivity claims have some truth - but with serious caveats. The AI assistants made both brilliant architectural decisions and bizarre errors. They proved theorems for me in 30 seconds that would've taken me hours, yet also confused left with right in a simple data structure.

Still: this isn't the death of programming, nor the advent of English-as-a-6th-generation-programming-language, but IMO a shift in how we handle uncertainty and ambiguity in software development. Instead of a rigid human-to-machine dialogue, it's now a probabilistic collaboration where the fuzziness moves from your head into the conversation itself.

I documented everything - all 300+ exchanges, the errors, the wins, and what this means for developers with different experience levels. Including from a psychological and emotional point of view.

So if you're curious about what "English as a programming language" actually looks like in practice, here is my full article (10k words):

https://marcobenedetti.substack.com/p/vibe-coding-as-a-codin...

Code repository: https://github.com/mabene/vibe

Happy to answer questions about the experience. For context, I hold a (pre-LLM) PhD in AI and currently work as a research advisor for the AI team of a large organization, so I've been watching this evolution from both practical and theoretical angles.