a short response to "how to ruin all of package management"

30 Dec, 2025

https://nesbitt.io/2025/12/27/how-to-ruin-all-of-package-management.html

in general I agree with Andrew's argument here -- the commodification of influence/centrality/importance metrics is going to utterly destroy their value. however, I'm less sold on the specific examples he puts forward of how AI, crypto, and prediction markets constitute a step change rather than another new frontier for the trends that already existed as a result of Good Old-Fashioned Capitalism.

Package metrics would make excellent prediction markets. Will lodash hit 50 million weekly downloads by March? Will the mass-deprecated package that broke the internet last month recover its dependents? What’s the over/under on GitHub stars for the hot new AI framework? These questions have answers that resolve to specific numbers on specific dates. That’s all a prediction market needs. Manifold already runs one on GitHub stars.

the Manifold market here is on the repo for the software that runs Manifold itself. it has a total depth of 2,225 MANA, which is supposedly "play money" with no monetary value -- the sidebar of the site offers you 1,000 MANA for signing up. as of when I started writing this, 98.7% of the depth of the market is staked on there being 500-750 stars (it's at 476, with two days left to go¹), but it dropped to 86% when apparently one user took a 20 MANA position against that happening.

The tea.xyz experiment

Tea.xyz promised to reward open source maintainers with cryptocurrency tokens based on their packages’ impact. The protocol tracked metrics like downloads and dependents, then distributed TEA tokens accordingly. [...]

the real innovation in scamming with TEA was that the aura of technical sophistication that surrounded cryptocurrency created a market of people with money who were willing to give money to an organization with a business plan which, when you stripped away that aura, was "we will give developers five cents for every GitHub star they earn, and this will also be profitable somehow for both us and you, our valued investor."

Andrew himself admits that crypto is not an enabling technology here later in the section:

Even well-intentioned open source funding efforts can fall into this trap. If grants or sustainability programs distribute money based on downloads or dependency counts, maintainers have an incentive to split their packages into many smaller ones that all depend on each other.

A developer might pause at a package with 10,000 stars but three commits and no issues. An AI agent running npm install won’t hesitate. It’s pattern-matching, not evaluating.

I'm not a "stochastic parrot" guy in general, but the developer is also matching a pattern when they pause, and "one integer-valued statistic is wildly disproportionate to two other integer-valued statistics" is an extremely straightforward pattern to match; you don't even need a neural network to do it. while I don't think AI agents would match it currently, there are plenty of real-world deployed systems which have freestanding fraud-detection models. and, of course, by Andrew's own argument, it is only by the grace of God that there isn't already a market for gaming these other metrics as well, to match the hero metric and allay suspicions -- blasting the commit graph with a bunch of indentation changes that cancel each other out, or issue reports that get bulk-resolved as duplicates...

anyway, back to the top. in general, I do agree with the thesis here: a lot of failures of the original promise of the internet qua public square are, I think, owed to the fact that everyone felt compelled to hang hard statistics and monetary rewards on the systems of human communication, and in so doing, created an opportunity for people to mass-generate a broad spectrum of degenerate (in the mathematical sense) uses of the medium which look statistically great and make money for a while but provide very little enduring value to people.

SEO, clickbait, ragebait, chumbox ads full of One Weird Tricks, listicles, vertical video with Subway Surfers gameplay footage... they've been with us for years and years. for sure, AI makes it possible to churn out a larger quantity of this stuff that looks less fake than the fake content of yesteryear -- which used to be churned out by botnets of compromised home PCs, or buildings full of impoverished people in developing countries, or just a tiny little shell script.

and one thing I think Andrew weirdly neglects to mention is that there's been a slow drip of high-profile compromised packages in the NPM ecosystem over the past few years; these packages generally perform tasks which are agnostic to product category, but the exploits themselves are almost universally crypto-stealers which target users with crypto wallets or software which processes cryptocurrency transactions. (the first incident that comes to mind.) the short line from arbitrary code execution to money appearing in your bank account does create a financial incentive to exploit packages, and an incentive to strike as low in the dependency tree as possible.

but earlier in 2025, something happened completely independent of AI, crypto, and prediction markets that could've dwarfed all of these frauds combined: Scattered Lapsus$ Hunters announced that they had hacked CrowdStrike, and a couple months later CrowdStrike announced that what had happened was that an insider had leaked screenshots of their internal dashboards as evidence of a fake compromise.

let's trust that CrowdStrike isn't just lying for a second. there's an easy way to get this access: find a disgruntled CrowdStrike employee; trade on your name as a member of a powerful and feared hacker group; offer to cut them in on a share of the proceeds for whatever scam you run in exchange for leaking you some screenshots.

once you have this access, there's a couple scams you can run. the simplest scam is to sell access through a darknet access broker, then mysteriously go missing when you're called upon to actually deliver it. (if this is even possible -- if I were dealing with a lot of people who are doing things that are on the wrong side of the law, I might try to set up some sort of escrow system.)

a more sophisticated scam? CrowdStrike is publicly traded; you can short CrowdStrike and then post to the access broker forum. after the market dips but before CrowdStrike can deny it, cover the short and leave with the proceeds.

#online #technology