Saturday, March 28, 2026

If you've been using Claude's desktop app, you might've noticed a .claude/ folder appearing in your projects. This breakdown explains what's actually in there — context files, project settings, and how Claude uses them to maintain state across sessions. Useful if you're trying to understand how to work with Claude's memory instead of against it. Discussion on HN.

Simon Willison got a new M5 MacBook Pro and decided to build custom Activity Monitor replacements using local LLMs and SwiftUI. The whole post is a nice case study in "vibe coding" — using AI to rapidly prototype GUI apps in unfamiliar frameworks. If you've been curious about building native Mac apps but never learned Swift, this is a convincing demo that the barrier is now much lower.

A team used AI to port JSONata (a JSON query language similar to jq) from JavaScript to Go in about a day, apparently saving half a million in annual costs. The framing is a bit hyperbolic, but it's another data point in the "vibe porting" trend — using LLMs to translate entire codebases between languages. The time savings here are wild if the numbers hold up.

Someone built a delightful Animal Crossing–inspired interface for working with Claude's coding agents. It's open source and actually looks fun to use — a reminder that developer tools don't have to be all terminal windows and JSON logs. Worth checking out if you're building anything with AI agents and want inspiration for better UX. HN thread.

Greg Kroah-Hartman, a Linux kernel maintainer, says AI-generated bug reports suddenly got useful — they went from noise to actually actionable. This is noteworthy because kernel maintainers have famously low tolerance for BS, so if they're saying AI tools crossed a quality threshold, that's a real signal. The article doesn't say which tools made the leap, but it's a useful marker for where we are.

Anthropic's Claude slipped below 99% uptime this quarter for the first time. It's a small data point, but worth noting if you're building anything mission-critical on top of it. Reliability is table stakes as these tools move from "cool demos" to "things people depend on." HN discussion.

STADLER, a manufacturing company founded in 1796, rolled out ChatGPT to 650 employees and is seeing measurable productivity gains. It's a case study, so take it with a grain of salt, but interesting to see a very old-school industrial company betting this hard on AI for knowledge work. The details on how they're actually using it are worth skimming.