<h2>Must know</h2>
<ul>
<li><strong>Arm launches AGI CPU for agentic data centers</strong>: Arm has unveiled its first production silicon, the AGI CPU, featuring up to 136 Neoverse V3 cores and 6TB memory support to power autonomous AI agents. According to reports from KubeCon, the chip utilizes a 3nm dual-chiplet design with PCIe Gen 6 and CXL 3.0 to double performance-per-rack over x86 systems. Launch partners including Meta, OpenAI, and Supermicro are already prepping liquid-cooled server configurations capable of housing 45,000 cores per rack. <a href="https://app.daily.dev/feed-by-ids?id=S5lBW0zAU&amp;id=7HHpd3WhV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI kills Sora to focus on the business super app</strong>: In a major strategic pivot, OpenAI is shutting down its Sora video generation tools across ChatGPT and its API effective immediately. CEO Sam Altman cited compute constraints and a need to eliminate ‘side quests’ in favor of core productivity and business tools. The company is now consolidating its desktop app, Codex, and browser into a single ‘super app’ powered by a new foundational model codenamed Spud. <a href="https://app.daily.dev/posts/55QkjBkbg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Massive supply chain attack hits Trivy and LiteLLM</strong>: A coordinated campaign by threat group TeamPCP has compromised the Trivy vulnerability scanner and LiteLLM packages, infecting over 1,000 cloud environments. Wiz Research reports the attackers exploited GitHub Actions misconfigurations to steal privileged tokens and inject credential-stealing malware that targets SSH keys and cloud secrets. Mandiant warns the victim count could reach 10,000 as the malware utilizes a novel ‘CanisterWorm’ to spread via npm and Docker Hub. <a href="https://app.daily.dev/feed-by-ids?id=FpC0RIGlf&amp;id=GU6PZXcza&amp;id=I6Q0A8ukt&amp;id=rQV4oxnV0" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>NVIDIA donates GPU resource driver to CNCF</strong>: The new Dynamic Resource Allocation driver replaces legacy models in Kubernetes, allowing workloads to request specific GPU topologies and NVLink-connected slices. <a href="https://app.daily.dev/posts/AoVkebmT8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic adds auto mode to Claude Code</strong>: A new research preview uses a classifier to autonomously approve safe tool calls while blocking destructive actions like mass file deletion. <a href="https://app.daily.dev/posts/Hgz3is1rZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vercel acquires new.website for v0 integration</strong>: The startup’s primitives for forms and SEO will be baked into v0 to enable agent-aware, production-ready website generation. <a href="https://app.daily.dev/feed-by-ids?id=TXo5OfLTR&amp;id=20Ih1mD20" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Research achieves 6x KV memory reduction</strong>: The new TurboQuant algorithm speeds up attention computation on H100s by 8x without sacrificing model accuracy. <a href="https://app.daily.dev/posts/9uUYicpAL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Optasia scales AI credit vetting to formal banking</strong>: The fintech is partnering with FirstRand Group to bring its low-default AI lending platform to underserved banking segments. <a href="https://app.daily.dev/posts/iLajreXbl" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Netflix Live Origin handles 65 million concurrents</strong>: The custom server architecture uses millisecond-grain caching and a Cassandra-backed store to manage 200 Gbps read throughput for live events. <a href="https://app.daily.dev/posts/yovKQxg1J" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>SentrySearch brings semantic search to dashcams</strong>: This open-source Python tool uses Gemini embeddings to search video footage natively without needing transcriptions or frame captioning. <a href="https://app.daily.dev/posts/TiSJPB0Uy" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The clean room AI license laundry</strong>: A service called Malice is claiming it can ‘launder’ GPL code by having one AI write a spec and another reimplement it, supposedly bypassing copyleft obligations. While they cite 40-year-old legal precedents like Baker v. Selden, the idea that a LLM’s ‘clean room’ isn’t just a high-speed plagiarism engine is a bold bet on a very thin legal fantasy. It’s less ‘innovation’ and more a commercialized middle finger to open-source sustainability. <a href="https://app.daily.dev/posts/BCnNy3tpC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA’s data center ghost towns</strong>: Investigative critiques are finally pointing out that while NVIDIA is printing money selling GPUs, only 33% of announced US data center capacity is actually under construction. We are effectively watching a ‘capital misallocation bubble’ where chips are being sold years before the buildings to house them—or the power grids to run them—even exist. It turns out you can’t actually run an AGI on a press release and a prayer. <a href="https://app.daily.dev/posts/X3my4Sf8s" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The 10x productivity mirage</strong>: Despite the hype from OpenAI and ‘vibe coders,’ a longitudinal study by DX found that AI tools are yielding a measly 10% productivity gain, not the promised 1000%. Management is still falling for the ‘tokens per developer’ fallacy, which is just the 2026 version of counting lines of code. If your PR review times have jumped 91% because of AI slop, you haven’t accelerated; you’ve just moved the bottleneck to the humans who have to fix the hallucinations. <a href="https://app.daily.dev/feed-by-ids?id=9ntgi4vJj&amp;id=RvQ2iIf5k&amp;id=zpUTGTPYX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
