<h2>Must know</h2>
<ul>
<li><strong>Anthropic adds auto mode to Claude Code</strong>: Anthropic has introduced a new permissions system for its CLI tool that uses a classifier to automatically approve safe tool calls while blocking destructive actions like mass deletions. This research preview for Team plan users aims to reduce prompt fatigue without resorting to the dangerous skip-permissions flag. It is currently compatible with the latest Claude 4.6 Sonnet and Opus models. <a href="https://app.daily.dev/posts/Hgz3is1rZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA and IBM donate AI infrastructure to CNCF</strong>: At KubeCon Europe, NVIDIA donated its Dynamic Resource Allocation driver to the CNCF to standardize GPU sharing and multi-node NVLink interconnects within Kubernetes. Simultaneously, IBM and Red Hat donated llm-d, a framework for distributed LLM inference that disaggregates prefill and decode phases to optimize hardware utilization. These moves signal a shift toward making high-performance AI workloads a first-class, open-source cloud-native citizen. <a href="https://app.daily.dev/feed-by-ids?id=AoVkebmT8&amp;id=tdsbXMmJT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Hypura enables massive LLM inference on base Macs</strong>: A new Rust-based scheduler called Hypura allows Apple Silicon users to run models that exceed their physical RAM by intelligently streaming tensors across GPU, RAM, and NVMe tiers. It specifically optimizes Mixture of Experts models like Mixtral by exploiting sparsity, achieving usable token speeds on hardware that would otherwise crash with out-of-memory errors. The tool exposes an Ollama-compatible API and handles prefetch depth automatically via hardware profiling. <a href="https://app.daily.dev/posts/9m4QpgqvK" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Snyk launches security agents for the MCP ecosystem</strong>: New tools automate AI Bill of Materials mapping and provide supply chain security for Model Context Protocol servers and agent skills. <a href="https://app.daily.dev/posts/iRGLDmExe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Astro 6 matches dev and production runtimes</strong>: The rewritten dev server now uses workerd to ensure local development perfectly mirrors Cloudflare Workers environments. <a href="https://app.daily.dev/posts/QG6z68w6t" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Haskell dataframe library hits version one</strong>: The release features a typed API that tracks schemas at compile time and adds an Apache Arrow interface for Polars interoperability. <a href="https://app.daily.dev/posts/jPEN94yLz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Epic Games cuts workforce by twenty percent</strong>: CEO Tim Sweeney announced layoffs of over 1,000 employees to pivot focus toward Unreal Engine 6 and Fortnite seasonal content. <a href="https://app.daily.dev/posts/G1SUSytCF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Slug GPU text rendering enters public domain</strong>: After a decade, the highly efficient Slug algorithm’s patent has been dedicated to the public, with HLSL shaders released under MIT. <a href="https://app.daily.dev/posts/EUFNfYm8U" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>SentrySearch brings semantic search to dashcams</strong>: This Python CLI uses Gemini Embedding 2 to index video chunks directly into ChromaDB for natural language footage retrieval. <a href="https://app.daily.dev/posts/TiSJPB0Uy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Node.js releases critical security patches</strong>: The 20.20.2 LTS update addresses seven CVEs, including timing-safe comparison flaws in Web Crypto and HTTP/2 flow control errors. <a href="https://app.daily.dev/posts/C8HkH2xnr" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI license laundering is the new clean room</strong>: A service called Malice is reportedly using AI to ‘clean room’ GPL code into MIT-licensed clones by having one LLM write specs and another implement them. While the legal theory is cute, it’s essentially a high-tech way to strip away copyleft protections and screw over maintainers who spent years on the original logic. The chardet library rewrite controversy proves this ‘laundering’ often results in ‘toxic’ code that enterprise legal teams won’t touch anyway. <a href="https://app.daily.dev/feed-by-ids?id=BCnNy3tpC&amp;id=G5hkjetWr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The modern data stack is collapsing into agents</strong>: MotherDuck predicts that by 2026, humans will barely write SQL and BI vendors will be irrelevant as ‘agent-driven architectures’ take over. It’s the classic ‘no-code’ dream rebranded for the LLM era, ignoring the fact that when the agent hallucinates a join, you still need a human who actually understands the schema to fix the mess. Expect a massive ‘text-to-SQL’ scandal once companies realize their autonomous agents are hallucinating financial metrics. <a href="https://app.daily.dev/posts/X1JY9WQX8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Fight AI with AI is pure garbage</strong>: Security vendors at RSAC are tripping over themselves to sell ‘AI-powered’ threat detection, but as ThreatLocker’s CEO points out, AI can’t actually determine software intent. Most of these tools are just adding latency and cost to problems that could be solved with basic Zero Trust and blocking untrusted binaries. We’re being sold expensive, probabilistic ‘AI firewalls’ to fix security holes that shouldn’t have existed in the first place. <a href="https://app.daily.dev/feed-by-ids?id=f6diiN9F1&amp;id=k3pmCJTXy" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
