The most-downloaded package on your dependency list can be compromised for three hours and the only reason anyone notices is a bug in the malware. @sdrzn, founder and CEO of Cline, builds a talk around that kind of failure: "Open Source Is Dead. Long Live Open Source." is on @aiDotEngineer's YouTube. It's an argument about which parts of open source are actually dying and which parts are about to matter more, with the economics laid out. - The community layer is what died. Zig's code of conduct bans AI on PRs, issues, and comments, because the core team values growing trusted contributors over the contributions themselves. Curl's CEO says AI-generated bug reports are effectively DDoSing the project. TLDraw now auto-closes pull requests. GitHub added a feature to disable third-party PRs altogether. - Supply chain risk is the other half. LiteLLM, at about 3.5 million downloads a day, was compromised for three hours: stolen PyPI publishing tokens, a credential harvester going after API, SSH, and crypto keys, plus remote command execution. It was caught by luck. - Inference spend is the pressure everyone's under. A CFO's anonymous report of $500M on Claude in a month after nobody set usage limits. Uber burning its entire 2026 budget in four months at $2,000 per user monthly. - The labs are subsidizing lock-in. Semianalysis found a $200 Claude plan yields roughly $8,000 of API usage and a $200 Codex plan about $14,000. Saoud reads that as building dependency ahead of the price rise. - Open weights are close enough for the money to move. Cline tested GLM against Opus on a real bug from their repo. GLM used twice the tokens at half the cost, cleaned up dead code, and verified the build compiled. Opus was faster but left type errors and broke the production build. - Intelligence belongs in the system, not just the model. Skills, rules, verification, quality gates. With those, a weaker model gets to the same place, just with more tokens. - Open Compute is the precedent. Facebook gave away its data center designs in 2011, the supply chain standardized on them, component prices collapsed, and Facebook saved billions on its own costs. - A closing ask to the American labs. Release more open weights, or the infrastructure gets built on foreign models and the lead goes with it. I'm working through the published talks from AI Engineer World's Fair sharing summaries and takeaways. Follow for more!