Vibe Coding Promises 10x Productivity Gains — PyPI Hasn't Confirmed It

Vibecoders and agentic tool enthusiasts say they've become 2x, 10x, even 100x more productive. One developer built a browser from scratch over the weekend. Skeptics ask a reasonable question: if this is true — where is all this new software?

The vibecodding phenomenon

Since 2023, AI assistants in development have transformed from laboratory experiments into everyday tools. Cursor, GitHub Copilot Workspace, Claude Code, Windsurf — new products launch every quarter. The principle is the same: a developer states an intention, a language model generates code, the developer reviews and refines it.

Claims about productivity sound serious. They come not from beginners, but from experienced engineers with years of experience: 5–10x speedups on routine tasks, an MVP in a week instead of a month, a SaaS product built solo in a month instead of a quarter by a team. On social media, this is no longer an exception — it's a genre.

Economic logic suggests: if development got cheaper — there should be more of it. Such questions stem from the assumption that the world wants more programs, and therefore if they're cheaper to build — they'll build more of them. If you agree with that, then there should be a measurable "AI effect."

The PyPI test

PyPI — the central repository for Python packages — is an ideal gauge for this hypothesis. It's large and stable, the data is public and collected over years. Python is one of the primary languages in the AI ecosystem, so if the effect manifests publicly, it would be here.

Signs we would expect to see after 2023:

• increase in the number of new packages published per month
• increase in the number of authors releasing a package for the first time
• emergence of new subject clusters (agents, LLM tools)
• acceleration of release cycles in existing libraries
• reduction in time from first commit to publication

PyPI data is open and available through BigQuery. If the publication growth curve sharply rises after 2023 — that's strong evidence for the vibecodding narrative. If not — the effect exists, but manifests differently than expected.

Why the data might be silent

Even if the growth is real, PyPI might not show it. Most vibecodding projects never reach public registries: personal utilities, company internal tools, prototypes for clients — all of it stays private.

AI accelerates the beginning first: sketch the architecture, generate the framework, write tests. The slow parts — final debugging, documentation, post-release support — remain slow. Result: projects start faster, but finish no more often.

There's also an ambition-growth effect. Before, a developer would spend a week on a simple script. Now with the same week, they take on a full product with UI, API, and database. The total volume of published code didn't double — but the complexity of each project increased. PyPI doesn't register this.

"If we don't see a doubling of package count, then the speed gains are

being absorbed by something else" — roughly how the skeptics' central thesis sounds.

What this means

Vibecodding is a real shift in how development feels. But public metrics so far don't confirm an explosive growth in the volume of software being created. Possible explanations: most AI projects remain private; the gains are redirected toward higher-complexity tasks; or we're at the beginning of the curve and data will emerge in a few years.

In any case, before taking 10x and 100x figures at face value, you should look not at tweets, but at repositories.