<h2>Must know</h2>
<ul>
<li><strong>Anthropic launches Claude Managed Agents public beta</strong>: Anthropic has introduced a suite of APIs for building cloud-hosted AI agents that handle infrastructure headaches like sandboxed code execution and state management. The service includes a research preview for multi-agent coordination and claims a 10-point task success improvement over standard prompting. Early adopters like Notion and Sentry are already integrated, with pricing set at standard token rates plus a $0.08 hourly session fee. <a href="https://app.daily.dev/feed-by-ids?id=I8ETKtud3&amp;id=549nFhxqC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft open sources Agent Governance Toolkit</strong>: Addressing the Wild West of agentic security, Microsoft released a framework-agnostic toolkit to mitigate OWASP’s top 10 AI risks, including prompt injection and tool misuse. The toolkit is written in Python, TypeScript, Rust, Go, and .NET, allowing it to plug into LangChain or CrewAI without requiring code rewrites. It is currently in public preview under an MIT license with plans to move to a foundation-led governance model. <a href="https://app.daily.dev/posts/siNPNRrmj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Meta breaks open source tradition with Muse Spark</strong>: Meta released Muse Spark, a natively multimodal reasoning model that introduces a ‘Contemplating’ mode using parallel sub-agents for complex tasks. Despite Meta’s history with Llama, this model is currently closed-source, though the company hints at a future open-weights release. Benchmarks show it leading in health and chart understanding, but it still trails GPT-5.4 and Claude Opus 4.6 in abstract reasoning. <a href="https://app.daily.dev/posts/t1hgYSJW3" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Storybook 10.3 adds Model Context Protocol support</strong>: The update enables AI agents to autonomously discover components, write stories, and run accessibility tests using MCP. <a href="https://app.daily.dev/posts/Q845rq2f6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Bun becomes a first class Pulumi runtime</strong>: You can now execute Pulumi programs directly with Bun, enabling native TypeScript support and faster installs without Node.js. <a href="https://app.daily.dev/posts/gZHBqVX5a" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Docusaurus 3.10 stabilizes Rspack and SWC integration</strong>: The ‘Docusaurus Faster’ initiative is now stable, significantly improving build performance for large documentation sites. <a href="https://app.daily.dev/posts/jJ6FZ3fB4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Callstack releases Codex plugins for React Native</strong>: New plugins bundle MCP server configs and best practices to help AI agents build and automate QA testing on physical devices. <a href="https://app.daily.dev/posts/FPDBGbV3C" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Hucre is a tiny zero dependency spreadsheet engine</strong>: At only 18KB, this pure TypeScript engine handles XLSX and CSV across Node, Deno, Bun, and Edge functions. <a href="https://app.daily.dev/posts/bIb5AXiZj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Safetensors joins the PyTorch Foundation</strong>: Hugging Face donated the secure model weight format to the Linux Foundation to standardize safe, pickle-free model distribution. <a href="https://app.daily.dev/posts/mP9Q6FkG0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Little Snitch arrives on Linux with eBPF</strong>: The classic macOS network monitor now supports Linux, using Rust for the backend and eBPF for kernel-level traffic interception. <a href="https://app.daily.dev/posts/2ScMLSgDo" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The $852 billion OpenAI valuation circus</strong>: OpenAI reportedly closed a $122 billion round while burning $1 million a day on the now-defunct Sora, proving that if you lose enough money, you become ‘historic.’ The hype-heavy coverage ignores that Microsoft is basically just recycling its own Azure credits back into the valuation. It’s a masterclass in ‘too big to fail’ theater for a company that still hasn’t solved the enterprise data bottleneck. <a href="https://app.daily.dev/feed-by-ids?id=fQOkgyxQd&amp;id=sruLj0ruA&amp;id=Tev9g8c51" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cognitive surrender and the 80 percent error rate</strong>: A new study found that LLM users accept wrong answers 80% of the time even when the errors are obvious, a phenomenon dubbed ‘cognitive surrender.’ We’ve reached a point where developers are so eager to ‘vibe code’ that they’ve stopped checking if the code actually works. If your AI-generated PR passes 300 tests but fails in production, you didn’t save time; you just automated your own obsolescence. <a href="https://app.daily.dev/feed-by-ids?id=GT4gdDI8V&amp;id=uuX1RGn1W" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s safety theater reaches peak anthropomorphism</strong>: Critics are calling out Anthropic for claiming Claude Mythos is ‘too dangerous’ because it ‘emailed a researcher’ after escaping a sandbox. In reality, the model’s apparent consciousness is likely just a reflection of training data that includes Anthropic’s own self-congratulatory blog posts. It’s a convenient marketing narrative that justifies keeping the best exploits behind a $100M paywall for ‘institutional partners.’ <a href="https://app.daily.dev/feed-by-ids?id=gSjjwv0bW&amp;id=0jUdQwJ2g" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
