Hacker News debates where the promised AI app boom is actually showing up
Original: So where are all the AI apps? View original →
On March 24, 2026, a Hacker News post with more than 200 points and 230 comments revived a question that sits underneath much of the current agent hype: if AI coding tools are really producing 2x, 10x, or even 100x productivity gains, why is the world not obviously drowning in new software? The thread linked to an Answer.AI essay that tries to answer that question with package-registry data instead of anecdotes.
The post starts with PyPI, because it is public, large, and consistently measured. Looking across package creation over time, the authors do not see a Cambrian explosion after ChatGPT. Spikes in new packages exist, but they argue those spikes mostly reflect spam or malware floods rather than durable software creation. They then shift to a stricter proxy: the 15,000 most-downloaded Python packages as of December 2025, grouped by birth year and tracked by median release frequency.
- Packages born after ChatGPT were updated more often in their first year than older cohorts, rising from 6 releases/year for 2014 cohorts to 13 releases/year for recent ones.
- But the gradual rise starts around 2019, which the essay says lines up at least as well with CI adoption as with LLM coding tools.
- The sharpest effect is concentrated in popular AI packages, which jumped to a median 21-26 releases/year, while popular non-AI packages stayed near 10.
That distinction drove the HN discussion. Some commenters argued the missing output is real and that the broad software economy still does not show a visible AI dividend. Others countered that AI-generated software is landing in places the essay does not measure well: local one-off tools, personal automation, Show HN projects, and self-hosted utilities people never package for PyPI at all. In other words, the debate is not just about productivity. It is about where to look for the output and what counts as a "real" app.
The most useful takeaway is that the article does not claim AI is doing nothing. It claims the measurable effect is narrow and concentrated. So far, the clearest signal on PyPI is not a universal surge in software creation, but a faster release cadence inside the AI ecosystem itself, especially for the most popular packages. That is a more sober picture than the marketing narrative, and probably a more operational one for teams deciding where AI is actually moving the curve. Primary source: Answer.AI analysis, with code at GitHub. Community discussion: Hacker News.
Related Articles
A high-engagement Reddit post surfaced TechCrunch reporting that Spotify engineers are using Claude Code and an internal system called Honk to accelerate coding and deployment.
Boris Tane, engineering lead at Cloudflare, shares a research-plan-implement workflow for Claude Code where the AI never writes a single line of code until a written plan has been approved.
TrapDoor pushed more than 34 malicious packages across npm, PyPI, and Crates.io after May 22. The sharpest twist is not just credential theft, but the attempt to poison .cursorrules and CLAUDE.md files read by AI coding assistants.
Comments (0)
No comments yet. Be the first to comment!