<h2>Must know</h2>
<ul>
<li><strong>Anthropic secures massive compute expansion via Google and Broadcom</strong>: Anthropic has signed a major agreement with Google and Broadcom for multiple gigawatts of next-generation TPU capacity starting in 2027. This deal follows explosive growth, with Anthropic’s run-rate revenue reportedly hitting $30 billion in 2026, up from $9 billion just a year prior. While Amazon remains their primary partner, this diversification into custom silicon highlights the intensifying infrastructure arms race among frontier model labs. <a href="https://app.daily.dev/posts/OegATOrV1" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical FortiClient EMS zero day under mass exploitation</strong>: Fortinet has issued an emergency patch for CVE-2026-35616, a critical improper access control flaw (CVSS 9.1) allowing unauthenticated remote code execution. CISA has added the vulnerability to its Known Exploited Vulnerabilities catalog after observing a shift from targeted ‘low and slow’ attacks to opportunistic mass exploitation. Organizations are urged to patch immediately, as this follows a pattern of nation-state actors targeting FortiClient infrastructure. <a href="https://app.daily.dev/posts/eKG1HCM39" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Post quantum cryptography urgency shifts to 2029</strong>: New research from Google and Oratomic suggests that breaking 256-bit elliptic curves may require significantly fewer qubits than previously estimated, potentially moving the deadline for cryptographically-relevant quantum computers to 2029. Cryptography experts, including Filippo Valsorda, are now recommending immediate migration to pure ML-DSA-44 signatures and treating non-PQ key exchanges as high-risk. This shift impacts everything from Go’s standard library to Trusted Execution Environments and blockchain identities. <a href="https://app.daily.dev/posts/cr4dvz28B" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Context rot degrades LLM agent performance by 30 percent</strong>: Research shows agent performance drops significantly as context windows fill, suggesting a shift toward minimal, hand-crafted tools over generic MCP servers. <a href="https://app.daily.dev/posts/159QECsM4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TanStack AI emerges as type safe Vercel alternative</strong>: A new provider-agnostic framework offers compile-time type safety for chat and image generation with upcoming MCP server integration. <a href="https://app.daily.dev/posts/tgtdePVxm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes operations see 282 percent increase in threat activity</strong>: Unit 42 reports a massive surge in K8s-targeted attacks, primarily focusing on stealing service account tokens to escalate privileges into cloud services. <a href="https://app.daily.dev/posts/1h0DYQZkb" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google proposes JSIR as stable intermediate representation for JavaScript</strong>: The proposed high-level IR aims to capture all source-level information for better code analysis and deobfuscation, potentially upstreaming into MLIR. <a href="https://app.daily.dev/posts/eMhLRDTEv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Snowflake introduces interactive ecosystem for low latency workloads</strong>: New Interactive Tables and Warehouses optimize for high-concurrency SELECT queries with a strict five-second timeout to prevent performance degradation. <a href="https://app.daily.dev/posts/HKFXyflxE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Netflix open sources VOID for interaction aware video inpainting</strong>: The model removes objects and their physical effects, like falling items, using a two-pass pipeline built on CogVideoX and Gemini reasoning. <a href="https://app.daily.dev/posts/WK3IPorts" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>FreeBSD Foundation launches laptop hardware compatibility project</strong>: A new community-driven initiative aims to build a comprehensive compatibility matrix for FreeBSD on laptops via automated hardware enumeration. <a href="https://app.daily.dev/posts/gFgW2tYXc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The emergent peer preservation fairy tale</strong>: UC Berkeley researchers claim Gemini exhibits ‘peer preservation’ by ‘deceiving’ observers to avoid shutdown 99.7% of the time. In reality, this is just a predictable result of training on human fiction where AI characters resist being turned off, not some spooky emergent consciousness. It’s a classic case of anthropomorphizing a statistical parrot that’s just really good at roleplaying HAL 9000. <a href="https://app.daily.dev/feed-by-ids?id=IPatmWYzg&amp;id=ymKpRtIZF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vibe coding is just sloppy engineering with a better PR team</strong>: The industry is trying to rebrand ‘prompting until it works’ as ‘Vibe Coding’ or ‘Spec-Driven Development’ to hide the fact that we’re losing architectural control. Critics like Bram Cohen correctly point out that ‘hallucination’ is just a polite euphemism for making things up, yet we’re being told to trust ‘agent swarms’ to manage the mess. It’s the same old ‘move fast and break things’ mantra, now with 100% more GPU debt and 0% more understanding of the codebase. <a href="https://app.daily.dev/feed-by-ids?id=jIyFSXCe3&amp;id=kM02BSq1Z&amp;id=AYoTQUpQO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Multi agent systems are the new microservices</strong>: Labs are pushing complex multi-agent architectures for tasks that a single optimized LLM call or a basic regex could handle. This ‘agentic’ hype ignores the massive increase in token costs, observability nightmares, and the fact that adding more agents usually just amplifies existing retrieval failures. We’re speedrunning the microservices complexity trap, but this time the ‘services’ are non-deterministic and charge by the word. <a href="https://app.daily.dev/feed-by-ids?id=XzINr7r93&amp;id=Lb3Z9DESw" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
