<h2>Must know</h2>
<ul>
<li><strong>AI code generation: faster, but at what cost?</strong>: AI code generation can speed up production by 55%, but it shifts bottlenecks to verification, increasing review times by 91% in high-AI-usage teams. Microsoft, for example, is seeing more emergency patches and attributes over 30% of its code to AI, raising concerns about quality. The DORA 2025 report confirms AI amplifies existing team dynamics: strong teams improve, struggling teams worsen, especially with legacy codebases where AI tools show minimal gains due to ‘context collapse.’ This suggests that while AI writes code faster, humans are still stuck cleaning up the mess, as reported by sources like The Register. <a href="https://app.daily.dev/feed-by-ids?id=CUMsxmpoK&amp;id=H1aWwdchu&amp;id=KYToiOjQs&amp;id=TCb4dVj23" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI security risks: exposed models and critical vulnerabilities</strong>: Researchers found over 175,000 Ollama AI hosts exposed online without safety guardrails, enabling potential resource hijacking and remote attacks. A critical 1-click RCE vulnerability was also discovered in OpenClaw, an AI assistant used by over 100,000 developers, allowing authentication token theft and arbitrary command execution. Meanwhile, SpaceX updated Starlink’s privacy policy to use customer data for AI training, raising significant privacy concerns, as detailed by sources like The Hacker News. <a href="https://app.daily.dev/feed-by-ids?id=2SdHHyeRJ&amp;id=YUSdAM3D6&amp;id=d76FYhLFT&amp;id=iEIaVGrhd&amp;id=yXHHUa4yo&amp;id=4nlzVZMEL&amp;id=ALGXJz29U&amp;id=fuL6WfKHt&amp;id=oAoGgXAAn&amp;id=zgJdriYuf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes ecosystem sees major shifts and new tools</strong>: The Cloud Native ecosystem is evolving rapidly, with Ingress NGINX and Kubernetes Dashboard projects announcing their retirement. On the brighter side, a new open-source Kubernetes dashboard called Radar has emerged, built as a single Go binary with an embedded React frontend, offering real-time cluster visibility without cloud logins or agents. This aims to address limitations of existing tools like Lens and k9s, as highlighted in articles on Hacker News. <a href="https://app.daily.dev/feed-by-ids?id=EdFbm39PB&amp;id=DiEunA7dc&amp;id=a3wJg58hN" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>NVIDIA NVFP4 quantization released by Red Hat for LLMs</strong>: Red Hat has released NVFP4-quantized versions of popular LLMs, leveraging NVIDIA’s 4-bit floating-point format for efficient deployment with near-baseline accuracy, especially effective for frontier-scale models. <a href="https://app.daily.dev/posts/OPe22PWoI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Research quantifies AI agent scaling principles</strong>: Google Research evaluated 180 AI agent configurations, finding multi-agent coordination improves performance by 80.9% on parallel tasks but degrades it by 39-70% on sequential tasks, emphasizing architecture alignment. <a href="https://app.daily.dev/feed-by-ids?id=EORUl59BL&amp;id=fjOJw0ShS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>MLflow integrates new evaluation metrics for LLM agents and RAG</strong>: MLflow 3.9.0 now integrates over 50 evaluation metrics from DeepEval, RAGAS, and Arize Phoenix, providing a unified API to assess LLM agents and RAG systems for conversational quality, retrieval, hallucination, and safety. <a href="https://app.daily.dev/feed-by-ids?id=3L44pW4ot&amp;id=5ZJWT4v6q&amp;id=JVTlCrLRm&amp;id=h7CGoaIP0&amp;id=zVcgznCs0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Wix introduces xEngineer model for AI-driven development</strong>: Wix launched its ‘xEngineer’ model, integrating AI to empower engineers with broader ownership across the development lifecycle, aiming to evolve responsibilities rather than reduce headcount. <a href="https://app.daily.dev/posts/1ESLU5xBs" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>VisualJJ extension brings Jujutsu version control to VS Code</strong>: VisualJJ is a new VS Code extension offering an interactive change tree, drag-and-drop commit rebasing, and integrated GitHub pull request management for the Jujutsu version control system. <a href="https://app.daily.dev/posts/uiLuflowX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>DataDog releases pg_tracing for PostgreSQL distributed tracing</strong>: DataDog developed <code>pg_tracing</code>, a PostgreSQL extension for server-side distributed tracing that generates spans for database queries and operations, supporting context propagation and OTLP export. <a href="https://app.daily.dev/posts/Xfr0u4GSj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>WebAssembly celebrates 10 years, expands beyond browser</strong>: WebAssembly (WASM) marked its 10th anniversary, evolving into a foundational technology for edge platforms, serverless runtimes, and embedded systems, offering deterministic, compact, and portable modules. <a href="https://app.daily.dev/posts/XjOBMuNCm" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Native PHP promises free mobile development with AI magic</strong>: Native PHP just dropped some ‘major updates’ making mobile development ‘completely free and open source,’ complete with an AI tool named Mimi that ‘rapidly generates complete Laravel mobile applications’ from natural language. Because apparently, writing actual code is for suckers when you can just wave your hands and have Mimi conjure up an app. What could possibly go wrong with AI-generated, ‘completely free’ mobile apps? <a href="https://app.daily.dev/posts/bRf9BXT9O" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI’s ChatGPT: web search is a feature, not a bug (unless it costs money)</strong>: OpenAI makes it ‘difficult’ to enable web search in ChatGPT, despite the model frequently hallucinating without it. The UX actively discourages search, leading one to wonder if OpenAI is more concerned with saving a buck on compute costs than providing accurate information. Who needs facts when you have a chatbot that confidently makes things up? <a href="https://app.daily.dev/feed-by-ids?id=0hvZGEXEr&amp;id=vgyq27hp2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI is killing productivity, but platform engineering will save us</strong>: Apparently, AI tools accelerate code generation but ‘paradoxically reduce overall productivity’ by enabling ‘cheap complexity’ and ‘organizational chaos.’ The solution? ‘Platform engineering’ and ‘golden paths’ to constrain AI-generated code. So, we’re building AI to write code faster, then building more layers of abstraction and control to rein in the AI’s ‘chaos.’ Sounds like a perfectly efficient plan, as noted by the article ‘Is AI killing productivity?’. <a href="https://app.daily.dev/posts/k9rFokMvG" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
