🔗 Interesting Links

Addy's key insight: the hard problem is coordination, not generation. Most teams don't need hundreds of agents — they need a modest set that solve real problems while maintaining control. He spends 30-40% of his time planning/specing before AI touches code. Also valuable: the distinction between feeling productive (100 agents running!) vs actually being productive (what did you build? is it making money?). The "planning is the new coding" shift is particularly relevant for code review — it's about knowing what good looks like before the AI gets to work.

Angular's new AI Tutor takes a project-based approach to learning, guiding you through building a "Smart Recipe Box" application. It uses an MCP (Model Context Protocol) server that integrates with AI-powered editors. The learning cycle is refreshing: brief concept explanations, generic code examples, then open-ended exercises where you solve problems yourself. The tutor reads your actual project files to verify your solution, and you can ask for hints or detailed guidance when stuck. It's a thoughtful implementation of AI-assisted learning that prioritizes critical thinking over passive code consumption - exactly the kind of interactive education that makes framework adoption less intimidating for developers returning after a hiatus.

Dan Abramov proposes treating social content like files - portable, app-agnostic, owned by the creator rather than the platform. He imagines syntax like `aliceowns` for ownership and `repost` for sharing. The core problem: platforms like Instagram, Twitter, and TikTok have locked our content inside walled gardens. If a platform dies or changes rules, our content might vanish or become inaccessible. A social filesystem returns ownership to users, but faces massive resistance from platforms with no incentive to build it. The decentralized, user-owned internet sounds ideal - the question is whether the next generation of social apps will make it reality.

Clawdbot is an open-source personal AI assistant that lives on your own infrastructure - whether that's a Raspberry Pi, Mac, or Linux box. Unlike cloud-based assistants, you control everything: it connects to WhatsApp, Telegram, iMessage, Discord, and more. It can manage emails, calendars, control smart homes, write code, and even modify itself through a hackable skill system. People are using it to check in for flights, build websites from their phones, and automate their entire digital lives. The testimonials are wild - someone built a whole website on a Nokia 3310 through it, another had it automatically control their air purifier based on biomarker goals. It's basically what Siri was supposed to be, but actually works.