The article measures the wrong thing. PyPI package creation is a terrible proxy for AI-assisted software output because packages are published for reuse by others, which requires documentation, API design, and maintenance commitments that AI doesn't help with much.

The real output is happening in private repos, internal tools, and single-purpose apps that never get published anywhere. I've been building a writing app as a side project. AI got me from zero to a working PWA with offline support, Stripe integration, and 56 SEO landing pages in about 6 weeks of part-time work. Pre-AI that's easily a 6-month project for one person.

But I'm never going to publish it as a PyPI package. It's a deployed web app. The productivity gain is real, it just doesn't show up in the datasets this article is looking at.

The iOS App Store submission data (24% increase) that someone linked in the comments is a much better signal. That's where the output is actually landing.