<h2>Must know</h2>
<ul>
<li><strong>Google’s embedding model deprecation causes RAG system failure</strong>: Google deprecated its <code>text-embedding-004</code> model, rendering existing vectors unusable for a production RAG system. This highlights a critical vendor lock-in issue where persisted embeddings become coupled to specific model geometries, making migration difficult. The solution involved re-embedding the entire dataset and a blue/green deployment, underscoring the need to treat embedding models as infrastructure and design for change, as reported by the author. <a href="https://app.daily.dev/posts/xhDI1zj4i" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Databricks launches Lakebase for AI workloads</strong>: Databricks launched Lakebase, a new PostgreSQL-compatible database specifically designed for AI workloads. It aims to bridge operational database capabilities with their data lakehouse architecture, offering a serverless OLTP solution with features like instant data branching and pgvector support. This move signifies a growing trend towards specialized databases optimized for AI, integrating transactional data directly with analytics platforms. <a href="https://app.daily.dev/posts/56RHAShlB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux kernel 7.0 rc1 released with Rust support</strong>: The Linux kernel 7.0 rc1 was released, with Linus Torvalds humorously attributing the major version bump to his dislike of large numbers, rather than a significant feature milestone. Key updates include cemented Rust support, faster cache clearing, non-disruptive kernel updates, and improved AMD/Intel silicon support. Torvalds also mused about future maintainer succession, referencing the community’s recently developed plan. <a href="https://app.daily.dev/posts/cYP8biu57" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>OpenAI’s Codex team leverages AI for internal workflows</strong>: OpenAI’s 40-person Codex team operates with minimal hierarchy and deep AI integration for onboarding, PR reviews, testing, refactors, and planning, favoring generalists or unconventional deep domain experts. <a href="https://app.daily.dev/posts/8TOu6Vf1U" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Developer productivity lost to non-coding tasks</strong>: A developer’s 10-second code fix for <code>flash-attention</code> took over 10 hours due to non-coding tasks like fighting sandboxed environments, <code>CUDA</code> conflicts, and build system issues, highlighting that <code>LLM tools</code> currently don’t address these major productivity sinks. <a href="https://app.daily.dev/posts/OBIqfWI7x" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Shuru: local-first microVM sandbox for AI agents on macOS</strong>: Shuru is a new local-first microVM sandbox for macOS AI agent workloads, using Apple’s Virtualization.framework to provide secure, ephemeral, and reproducible Linux environments without Docker. <a href="https://app.daily.dev/posts/TqgsBjEoV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes 1.35 enhances AI/ML workload support</strong>: Kubernetes v1.35 introduces features like workload-aware scheduling with gang scheduling support and stable in-place Pod resource resize, strengthening its role as an operational substrate for AI/ML workloads. <a href="https://app.daily.dev/posts/mAyDHBOpw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Intel’s OpenVINO 2026 improves NPU and LLM support</strong>: Intel released OpenVINO 2026, its open-source AI inference toolkit, with improved NPU handling and expanded support for large language models. <a href="https://app.daily.dev/posts/IGizSZZmf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ChatDetector uses LLMs to find RM-API misuse</strong>: ChatDetector is an LLM-powered system designed to detect resource management API misuse in software, using a ReAct-inspired framework and two-dimensional prompting to identify security bugs with high precision. <a href="https://app.daily.dev/posts/UGo4Ea1t8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>MySQL community calls for independent foundation</strong>: The MySQL community is urging Oracle to establish a vendor-neutral foundation for MySQL, citing declining popularity, missing modern features like native vector search for AI, and closed-door development. <a href="https://app.daily.dev/posts/UyLIHBdUw" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic’s Claude compiler video: reality vs. hype</strong>: Anthropic released a marketing video showcasing Claude autonomously building a C compiler capable of compiling Linux, SQLite, Lua, and Doom. Critics, however, claim the video used “dishonest framing,” suggesting the actual goal was a stress test of Claude’s extended autonomous operation, not a straightforward demonstration of compiler-building prowess. So, another day, another AI demo that needs a decoder ring to understand what actually happened. <a href="https://app.daily.dev/posts/vGYk0DbTO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Neiry tests ‘bio-drones’ with neural interface pigeons</strong>: Neiry, a deeptech company, announced early real-world testing of ‘bio-drones’—live pigeons fitted with neural interfaces, solar panels, and cameras to guide them along preset routes. Neiry claims these birds can cover 400 km daily for infrastructure inspection and search-and-rescue, with “privacy compliance” through on-device filtering. Because nothing says “cutting-edge tech” like strapping a brain-computer interface to a bird and calling it a drone, all while promising privacy with a straight face. <a href="https://app.daily.dev/posts/m7v8MA0Dh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Experienced developer falls for X phishing scam</strong>: A developer with 20 years of experience on X lost access to their account due to a sophisticated phishing attack, detailing five key mistakes made in a rush. The takeaways include verifying senders, scrutinizing URLs, and using authenticator apps over SMS 2FA. It’s a classic tale of “even the pros get phished,” proving that no amount of experience can save you from a moment of distraction and a well-crafted scam. Maybe AI can fix human gullibility next? <a href="https://app.daily.dev/posts/cFsXn3pdx" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
