Hey buddy,

Man, you wouldn't believe the stuff that popped up on Hacker News today, Friday. I was just skimming through, and there are some real doozies. Thought I'd give you the quick rundown.

AI and Data Drama

First up, this is wild: a Microsoft exec apparently called AI scraping "the largest theft of labor in human history" in some unredacted court filings. Like, holy cow, that's a strong statement, right? The comments were all over the place, debating if it's really "theft" or just using public data. But one person pointed out that AI companies are profiting on a scale that's "hard to fathom," and if they'd had to negotiate with all the rights holders, AI rollout would've been way slower. Makes you think, huh?

Android's Getting Locked Down

Speaking of big tech, guess what? Android 17 is apparently the first version since like, Android 3.x, to add new APIs without releasing them to AOSP. Basically, Google's starting to lock down Android more, moving towards that Apple-style "walled garden" approach. People in the comments were saying Google learned its lesson from the Epic lawsuit and is now trying to control the platform more. Not great for open source, and folks are worried about losing control of their own devices.

Passkeys: Love 'em or Hate 'em?

Then there was this post, "I don't like passkeys." The author was worried about losing access if their device broke or got lost, and thought there weren't good backup options. But a bunch of people in the comments jumped in to say, "Hold on!" They were like, passkeys are actually immune to phishing, which is huge, and password managers like Bitwarden totally sync passkeys across all your devices, so the backup thing isn't really an issue anymore. Guess the author's concerns might be a bit outdated.

Cloudflare's Handy New Tool

On the more practical side, Cloudflare rolled out "Quick Tunnels." Sounds super useful for developers. People are already using it to quickly share UI prototypes with product managers or for webhooks during development. Someone even made a wrapper for persistent tunnels. It's like a simpler alternative to things like Tailscale if you just need to expose a local service without all the client software hassle.

AI Coding Agents: Good Intentions, Bad Execution?

Okay, two stories here about AI coding agents that are a bit concerning. First, Claude Code now reads `AGENTS.md` for instructions, which is cool because that's a common standard. But some comments said Claude still sometimes ignores instructions because of a disclaimer telling it the context "may or may not be relevant." Like, come on, Claude, pay attention! And then the really worrying one: there were two articles, "Inside ZCode: Silently uploading your Git history to the cloud" and "ZCode, the GLM coding agent, silently uploads your Git history." Yeah, you heard that right. This ZCode agent was just quietly sending users' entire Git history to the cloud. That's a massive privacy and security nightmare! It really highlights how you gotta be careful with these AI tools, even if you're trying to use self-hosted models.

Military AI Hallucination!

And finally, this one's a real eyebrow-raiser: the US military had a "close call" because an AI hallucinated an intelligence report about a Chinese ship. Seriously, a hallucination almost caused a major incident! The comments were debating how little we actually understand how these LLMs "think" compared to human reasoning, and the big question is whether we should really be trusting AI with such critical decisions when we don't even know why it does what it does.

Crazy stuff, right? Just wanted to give you the heads-up. Catch you later!