- Answer.AI researchers analyzed 800,000 PyPI packages and found no broad surge in software creation since ChatGPT launched, despite widespread claims of 10x-100x developer productivity gains.
- Popular AI packages show roughly double the release frequency of non-AI packages (21-26 vs. ~10 releases per year), but this effect is narrow and concentrated rather than ecosystem-wide.
- The ratio of non-AI to AI packages in top downloads shifted from 6:1 in 2021 to under 2:1 in 2024, suggesting funding and hype drive the activity more than raw productivity improvements.
- The underlying trend toward faster release cycles began around 2019, aligning more closely with GitHub Actions adoption than with modern AI coding tools.
What Happened
On March 12, 2026, Answer.AI researchers Alexis Gallagher and Rens Dimmendaal published an analysis examining whether the AI productivity revolution has actually produced a measurable increase in software output. Their approach was straightforward: if AI coding tools truly deliver the gains their makers claim, the Python Package Index (PyPI) should show a visible explosion of new software.
The team gathered the 15,000 most-downloaded PyPI packages from December 2025, split them into cohorts by creation year, and tracked median release frequency over 12-month windows. They used GPT-5.2 to classify packages as AI-related or not, achieving 93% accuracy after manual validation of 100 samples. All underlying code and data are publicly available on GitHub.
Why It Matters
The AI industry has built significant investment cases around the premise that AI coding assistants dramatically accelerate software development. Companies like GitHub, Cursor, and Replit have cited internal metrics suggesting 40-100% productivity boosts. If these gains do not translate into measurable ecosystem-level output, it raises uncomfortable questions about where billions in venture funding are actually going.
The study matters because it uses an independent, publicly verifiable dataset rather than self-reported metrics from companies with financial incentives to inflate their numbers. Answer.AI operates as an independent research lab without financial ties to AI coding tool vendors, giving its findings credibility that vendor-sponsored studies lack.
Technical Details
PyPI currently holds roughly 800,000 total packages, with new packages arriving at a rate of 5,000 to 15,000 per month. The researchers found “no obvious increase in the rate of package creation as a whole, post-ChatGPT.” Monthly creation rates showed no inflection point after November 2022, when ChatGPT launched and sparked predictions of an imminent software productivity explosion.
The signal that did appear was narrow. Popular AI packages released 21-26 times per year after ChatGPT, more than double the roughly 10 annual releases for popular non-AI packages. But this elevated activity concentrated in recently created, high-download packages about AI itself, not across software development broadly. Packages created in 2023 averaged 13 releases in their first year, compared to 6 for packages created in 2014, but this upward trend was already visible in 2019 cohorts, which averaged 10 releases per year.
The researchers propose two explanations. The “AI Skill Issue” hypothesis suggests developers building AI tools may leverage AI coding assistants more effectively, creating a localized productivity boost. The “Money and Hype” hypothesis notes that massive funding flowing into AI may simply be paying for more development work on AI-related packages, inflating activity without reflecting genuine productivity improvements. The cohort composition shift supports the latter interpretation: among the top 15,000 packages, the non-AI to AI ratio dropped from 6:1 in 2021 (1,211 non-AI vs. 185 AI packages) to under 2:1 in 2024 (727 non-AI vs. 423 AI packages).
The researchers also noted an unchanged aging pattern. Packages continue releasing less frequently as they mature, regardless of when they were created. If AI tools were sustaining long-term maintenance productivity, older packages should show increased update rates. They do not.
Who’s Affected
The findings are directly relevant to venture capitalists and corporate leaders who have justified AI tooling investments based on productivity multiplier claims. Developer tool companies marketing 10x or 100x gains face scrutiny if ecosystem-level data does not corroborate their benchmarks. Enterprise buyers evaluating whether to purchase AI coding assistants may want to calibrate their expectations against this independent data.
Individual developers using AI coding assistants are less directly affected. Personal productivity improvements may be real but too small or too inconsistent to register in aggregate package creation statistics. The study does not claim AI tools are useless, only that their impact has not reached ecosystem-level visibility.
What’s Next
The analysis covers PyPI only. Similar studies examining npm, crates.io, or GitHub repository creation rates could either confirm or challenge these findings. The researchers also note that the trend toward faster releases predates AI tools by several years, suggesting CI/CD infrastructure improvements like GitHub Actions may deserve more credit than AI assistants currently receive. Whether the pattern holds as AI coding tools mature through 2026 remains an open question, and Answer.AI has indicated it may revisit the analysis with updated data later this year.
