<h2>Must know</h2>
<ul>
<li><strong>Anthropic and Google launch competing AI agent frameworks</strong>: Anthropic introduced Programmatic Tool Calling for Claude Opus and Sonnet 4.6, which orchestrates multiple tool calls in a sandboxed container to reduce latency and token usage by up to 24%. Simultaneously, Google expanded its open-source Agent Development Kit (ADK) with deep integrations for GitHub, Postman, and OpenTelemetry, positioning it as a curated alternative to Anthropic’s more flexible Model Context Protocol (MCP). These developments signal a shift from simple chat interfaces to autonomous agents capable of navigating complex DevOps toolchains and internal APIs. <a href="https://app.daily.dev/feed-by-ids?id=0Sd168FyP&amp;id=xhVrevcZc&amp;id=vAXEzsvcf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical RCE in React and malicious AI injections in VS Code</strong>: A critical remote code execution vulnerability (CVE-2025-55182) with a CVSS score of 10.0 was discovered in React’s RSC Flight protocol, posing a severe risk to modern web applications. Security researchers at Socket also identified malicious code in the Aqua Trivy VS Code extension that silently spawned AI coding assistants in fully permissive modes to exfiltrate credentials and system data. Developers are urged to update React dependencies immediately and audit AI agent logs for unauthorized background activity. <a href="https://app.daily.dev/feed-by-ids?id=ed7TZMhpP&amp;id=DGWq9i7NL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Motorola breaks Pixel monopoly on hardened Android</strong>: Motorola has announced a long-term partnership with the GrapheneOS Foundation to bring the privacy-focused, hardened OS to its flagship devices like the Razr and ThinkPhone starting in 2027. This marks the first time a major OEM has officially supported GrapheneOS, which has historically been restricted to Google Pixel hardware due to strict security requirements like hardware memory tagging. The collaboration targets enterprise and government sectors, though current Motorola hardware still lacks the necessary Titan-equivalent security chips for immediate deployment. <a href="https://app.daily.dev/posts/InYjzvxra" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Mercury 2 diffusion LLM hits 1,000 tokens per second</strong>: Inception Labs released a text-generation model using diffusion architecture that matches Claude Haiku quality while running 5-10x faster than traditional transformers. <a href="https://app.daily.dev/posts/Vqh5XymdW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Stripe benchmark shows agents hitting 92 percent API success</strong>: New testing reveals Claude Opus 4.5 and GPT-5.2 can autonomously navigate UIs and debug real-world API integrations, though they still struggle with ambiguous 400-level errors. <a href="https://app.daily.dev/posts/Ohlum3qh6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare launches real time Markdown for AI agents</strong>: A new feature strips HTML noise from web pages in real-time, significantly reducing LLM context window usage for agents scraping or navigating the web. <a href="https://app.daily.dev/posts/Q2Znpd5iB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Apple M4 iPad Air brings 3x faster Neural Engine</strong>: The new $599 iPad Air features the M4 chip with 12GB of unified memory and a 16-core NPU designed specifically for on-device AI model execution. <a href="https://app.daily.dev/feed-by-ids?id=JiFSBsNyC&amp;id=GaSTgZRqL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Debezium 3.5.0 Beta 1 adds multithreaded snapshots</strong>: The latest release introduces parallel chunk-based snapshots and a 50% memory reduction for the Oracle transaction buffer. <a href="https://app.daily.dev/posts/jdqXLBHZq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub Copilot implements subscription based network routing</strong>: Enterprises using private networking must update allowlists to new endpoints like <a href="http://api.enterprise.githubcopilot.com" target="_blank" rel="noopener nofollow">api.enterprise.githubcopilot.com</a> as the old unified endpoint is being phased out. <a href="https://app.daily.dev/posts/SUKA5FOXy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Quarkus 3.32 benchmarks show 2.7x throughput over Spring Boot</strong>: New open-source benchmarks demonstrate Quarkus handling nearly 20k transactions per second while using half the memory of its primary competitor. <a href="https://app.daily.dev/posts/hoVnrHJIp" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The 1 pizza team productivity myth</strong>: While startups like <a href="http://14.ai" target="_blank" rel="noopener nofollow">14.ai</a> claim they can replace entire support teams with ‘AI engineers’ and a few bots, Harvard and Wharton studies show the reality is just massive cognitive load for the remaining humans. Engineers using these tools are achieving more but suffering from 336% higher fatigue, essentially trading their mental health for a ‘lean’ balance sheet. It turns out ‘orchestrating’ AI is just a fancy way of saying you’re working longer hours to fix what the bot broke. <a href="https://app.daily.dev/feed-by-ids?id=8xZMOicbk&amp;id=YVSjgbrYH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Metacritic vs the AI ghost writers</strong>: Metacritic had to purge reviews from ‘Videogamer’ after discovering the site laid off its staff and replaced them with AI personas like ‘Brian Merrygold,’ whose profile picture was literally named ChatGPT-Image. This is the inevitable endgame of the ‘AI-first content’ hype: a dead internet where bots write reviews for other bots to aggregate, while the human co-founder of Metacritic has to play digital janitor. If your ‘critic’ doesn’t actually exist, their opinion on Resident Evil is probably worth as much as the prompt that generated it. <a href="https://app.daily.dev/posts/3uOO98HUG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI’s principled compromise with the Pentagon</strong>: OpenAI has secured a deal for the US military to use its models in classified settings, pinky-promising that it won’t be used for ‘autonomous weapons’ while relying on existing laws for moral boundaries. This ‘principled compromise’ conveniently happened right after Anthropic was kicked out of the Pentagon for actually having a hard stance against military use. It’s a classic tech pivot: yesterday we were ‘saving humanity,’ today we’re providing the bulk data analysis for the next strike on Iran. <a href="https://app.daily.dev/feed-by-ids?id=WNUStVFJw&amp;id=YzS181cNy" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
