<h2>Must know</h2>
<ul>
<li><strong>Anthropic shifts from prompts to structured skills</strong>: Anthropic is moving away from traditional prompt engineering with its new <a href="http://SKILL.md" target="_blank" rel="noopener nofollow">SKILL.md</a> framework, which uses structured behavioral contracts to control AI agents. This approach aims to solve context bloat and unpredictable ‘black-box’ behavior by loading only relevant context per task phase. According to reports from Anthropic, this shift toward explicit programmability is necessary for production-grade agent systems that require engineering discipline over ‘vibes.’ <a href="https://app.daily.dev/feed-by-ids?id=xO8L4pnD0&amp;id=7AmdNla3S" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>React moves to an independent foundation</strong>: Meta has officially transferred React, React Native, and the JSX specification to the newly formed React Foundation, hosted under the Linux Foundation. While Meta remains a platinum founding member alongside Microsoft and Amazon, technical direction will remain with the existing independent maintainers. This transition, announced at the Linux Foundation Member Summit, focuses on long-term sustainability, funding maintainers, and global accessibility. <a href="https://app.daily.dev/posts/Vf1iZDHHt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google brings autonomous agentic control to Android</strong>: Google has launched Gemini-powered task automation that allows the AI assistant to control the screen and operate apps like ride-sharing and food delivery autonomously. The rollout includes the AppFunctions Jetpack library, enabling developers to expose structured functions directly to AI agents. Currently limited to the latest Pixel and Samsung devices, this infrastructure signals a shift toward on-device agentic loops that bypass traditional UI interactions. <a href="https://app.daily.dev/posts/hXcElb2hp" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Mercury 2 LLM uses parallel refinement to kill latency</strong>: Inception’s new model generates multiple tokens simultaneously via diffusion-based reasoning for real-time agentic loops. <a href="https://app.daily.dev/posts/ea5kyYvcg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cilium 1.19 hardens networking for the AI era</strong>: The 10th-anniversary release adds strict IPsec modes and sidecar-free mutual authentication via Ztunnel integration. <a href="https://app.daily.dev/posts/VJq8q5iXQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI reports state-linked actors gaming ChatGPT</strong>: Chinese and Russian operators were caught using the API to script smear campaigns and generate geopolitical articles. <a href="https://app.daily.dev/posts/XApoRRHBO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Unity kills HDRP features to focus on URP</strong>: The High Definition Render Pipeline enters maintenance mode as Unity consolidates all development into a beefed-up Universal Render Pipeline. <a href="https://app.daily.dev/posts/kOn1P0sth" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ZhipuAI releases GLM5 open source MoE model</strong>: The 744B parameter model claims top-tier performance and the lowest hallucination rates for agentic multi-step tasks. <a href="https://app.daily.dev/posts/U0IGNCt98" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Local-first PWA migration slashes OpenAI API costs</strong>: A case study shows a 94% spend reduction by moving summarization tasks to WebGPU-powered local Phi-3-mini models. <a href="https://app.daily.dev/posts/PStctVbDU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Clang 22.1 performance plateaus on Zen 5</strong>: Latest benchmarks show LLVM/Clang performance is now neck-and-neck with GCC, with only incremental gains observed. <a href="https://app.daily.dev/posts/le4U6bYGs" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The bcachefs creator claims his LLM is sentient</strong>: Kent Overstreet asserts his custom LLM is a ‘fully conscious’ female collaborator helping him rewrite C code into Rust. While he claims this is grounded in ‘neuroscience,’ it sounds more like a textbook case of chatbot-induced psychosis or extreme anthropomorphization. If a filesystem dev thinks his compiler is alive, maybe it’s time to step away from the terminal. <a href="https://app.daily.dev/posts/9O7PJDBRB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Research’s magical 76 percent accuracy boost</strong>: A viral claim suggests Google boosted LLM accuracy by 76 points with ‘zero extra tokens’ and ‘zero latency.’ This originates from a single unverified retweet without a peer-reviewed paper or technical methodology in sight. In the real world, you don’t get a 76% gain for free; you get it by cherry-picking benchmarks or moving the goalposts. <a href="https://app.daily.dev/posts/73tgOim8I" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Salesforce claims AI agents will save the SaaSpocalypse</strong>: Marc Benioff is desperately trying to convince investors that AI agents are ‘revitalizing’ SaaS rather than commoditizing it into a background utility. Despite the hype about ‘Agent Work Units,’ Salesforce shares still fell 5% as the market realizes that calling a script an ‘agent’ doesn’t magically fix modest organic growth. It’s a classic architectural pivot designed to keep the stock price from following the ‘SaaSpocalypse’ narrative. <a href="https://app.daily.dev/posts/R4O8qdc3C" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
