Ron Northcutt

Fascinating: Switzerland has been quietly open-sourcing parts of its federal government’s digital infrastructure on GitHub 😳 The official “swiss” organization (run by the Federal Chancellery) hosts 55+ public repositories - design systems, API guidelines, admin tools, and knowledge standards - all available for inspection and reuse. While most governments keep their tech locked in black boxes, Switzerland treats open source as infrastructure. Most institutions still do this: ↳ Build in secret.
↳ Keep it closed.
↳ Pay vendors to repeat the cycle. Meanwhile, Switzerland is doing the opposite: turning code into something inspectable, improvable, and trustworthy. And that’s actually a brilliant strategy. Because when the tools become commoditized, the advantage belongs to whoever controls the source. P.S. check out How to Build an Agentic OS with Claude Fable 5 🤖: https://lnkd.in/dZsST5qR

Nobody is going to care how many AI seats you bought.   Your board will ask one question:   What changed?   Not how many people logged in. Not how many prompts ran. Not how many tokens got burned. Not whether the team “likes it.”   What changed?   Did a workflow get faster? Did a decision improve? Did a customer experience change? Did revenue move? Did margin improve?   That is the number. Everything else is theater.   The most common AI mistake I see right now is companies confusing activity with strategy.   They buy the licenses. Send the login. Run the workshop. Watch the invoice grow.   Then someone asks about ROI and the room starts telling stories about the future. Before you sign another AI contract, ask three questions:   What changed? What is it worth? What do we want next year?   I wrote the full breakdown on Substack. (link in first comment) 

Blue-green deployments not cutting it? Clover runs 3+ versions simultaneously using "rainbow deployments" with HAProxy One — routing each merchant to the right version via weighted map files. The result: daytime releases, zero on-call pages, zero drama🌈 https://hubs.la/Q04lT2JN0

Linus Torvalds on using AI tools in the Linux kernel. Full discussion: https://lnkd.in/gHkAxVUj 🐧 #Linux #OpenSource #AI #LLM

▌ The Registry as Business Model In January 2010 Isaac Schlueter released a small program for fetching JavaScript packages, and the thing it fetched them from was a registry that anyone could write to. That openness was the entire idea: you published and the world installed, with nobody at the door. Four years later there was a company around it, npm, Inc., and that openness was now the ground a business stood on. Sixth of eighteen Wednesdays, the first on npm. The honest place to start is the ledger that made publishing free. ■ THE FREE THING AND THE TILL The public registry costs real money to run and earns nothing directly. The revenue lived elsewhere: private packages, and an on-premises product. In spring 2015 the company raised eight million dollars and launched Private Modules at seven dollars per developer a month. That is freemium: the free thing draws the crowd, and its size makes the paid thing plausible. The asset is the crowd, measured in packages, and a company measured by how much it holds has little reason to hold less. ■ WHERE THE MONEY MEETS THE BOUNDARY The responsibility line, the place where a component's accountability ends, is simply absent from node_modules. That read as an architectural oversight. It reads at least as well as economics. A real boundary would cost gatekeeping and review at publish time, and friction shaves the growth of the count. A boundary keeps things out, which is an awkward device to fit to an inventory you are valued by the size of. So the line was never drawn, and the install script runs as you, with your keys and tokens, before you run a line yourself; one compromised maintainer reaches everything, the dividend of a model paid to ask nothing at the door. ■ WHAT CHANGED HANDS, AND WHAT DID NOT In 2018 Microsoft bought GitHub for seven and a half billion dollars, a sum that now looks unremarkable. In March 2020 GitHub, by then a Microsoft subsidiary, bought npm: over 1.3 million packages, seventy-five billion downloads a month. Microsoft paid for the size. The releases spoke of an end-to-end secured supply chain; the install script stayed where it had always been. The trust model outlived every owner, because it was never once the thing on sale. ■ THE PRICE OF A LINE A responsibility line is not a fantasy. libc reaches you with named maintainers and a signed, tested base: you do not read it, yet you know where it lives and who answers when it breaks. Someone is paid to stand at the door and say no, the expense a registry valued by its count would rather not carry. THE POINT The missing boundary was the business working as designed. Drawing it would have cost the asset the one number it was built to grow, and no owner since has found a reason to. The ungoverned middle held steady through three sets of hands, priced in from the start. Next Wednesday: the code quality question, with names attached. #npm #javascript #nodejs #softwaresupplychain #packagemanager

Most performance metrics reward the exact behavior they’re supposed to prevent. This meme reminded me of a few workplace classics. Recruiters measured on time-to-fill. They rush candidates through the process and make weaker hires. Developers measured on pull requests merged. Every feature gets split into five tiny PRs because quantity beats impact. Sales teams measured only on revenue. Heavy discounting and bad-fit customers suddenly become “wins.” Customer support measured on tickets closed. Quick fixes replace root-cause solutions because reopening a ticket hurts the metric. SREs measured on incidents closed. They become experts at resolving outages instead of preventing them. Consultants measured on billable hours. The fastest way to finish the work becomes the worst way to hit the target. People don’t optimize for outcomes. They optimize for whatever you measure.