<h2>Must know</h2>
<ul>
<li><strong>Apple M5 chips target AI performance amidst RAM supply crunch</strong>: Apple has unveiled the M5 chip family for MacBook Air and Pro, claiming up to 4x faster AI task performance over the M4 through a new Fusion Architecture that merges two dies. Despite the performance leap, MacBook Pro prices have climbed by up to $400, a hike analysts attribute to the global RAM shortage fueled by AI infrastructure demand. While the hardware supports Wi-Fi 7 and Thunderbolt 5, the Asahi Linux community notes that mainline kernel support for these new Pro/Max variants is likely years away. <a href="https://app.daily.dev/posts/FGAdSpUhd" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Chrome shifts to fortnightly releases to counter AI browsers</strong>: Starting September 2026 with version 153, Google Chrome will move from a four-week to a two-week release cadence for desktop and mobile. This accelerated schedule is a direct response to the rise of agentic web browsers like OpenAI’s ChatGPT Atlas and Perplexity’s Comet. While enterprise users can stick to an eight-week cycle, Google is pushing the faster track to ship performance fixes and web platform capabilities at a pace that matches the rapidly evolving AI landscape. <a href="https://app.daily.dev/posts/SAMWZskhL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>North Korean StegaBin campaign targets developer environments</strong>: The Lazarus-linked group Famous Chollima has released 26 malicious npm packages that typosquat popular libraries to deploy a multi-stage infostealer. Dubbed StegaBin, the attack uses Pastebin steganography to hide command-and-control infrastructure and specifically targets SSH keys, VS Code configurations, and cryptocurrency wallets. This evolution of their previous operations features advanced evasion like control flow flattening and Vercel routing to bypass traditional security scans. <a href="https://app.daily.dev/posts/QDsiIqiel" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Kubernetes v1.35 enables in-place pod resizing without restarts</strong>: The new stable feature allows updating CPU and memory limits on running containers by modifying cgroup limits directly via the CRI, eliminating the restart penalty for stateful apps. <a href="https://app.daily.dev/posts/fjhms6Zq4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kafka adds native queue semantics with Share Consumer API</strong>: KIP-932 is now generally available, allowing multiple consumers to process messages from a single topic without the traditional 1:1 partition-to-consumer constraint. <a href="https://app.daily.dev/feed-by-ids?id=AQzVX7AjG&amp;id=T0CcYg8af" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rust iterator abstraction causes 60x slowdown in Turbopuffer</strong>: Profiling revealed that recursive next calls in zero-cost iterators prevented compiler vectorization; switching to a batched iterator approach dropped query latency from 220ms to 47ms. <a href="https://app.daily.dev/posts/U3pwVYd72" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google and MIT release framework for scaling multi-agent systems</strong>: The predictive framework identifies a tool-coordination trade-off and claims 87% accuracy in determining whether a task requires centralized orchestration or decentralized navigation. <a href="https://app.daily.dev/posts/iN9oYGN2z" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Helm 4 adopts WebAssembly for plugin isolation and portability</strong>: The new Wasm-based system replaces the legacy Helm 3 model, offering sandboxed execution and up to 40% lower latency for Kubernetes extensibility. <a href="https://app.daily.dev/posts/W9pY3Qv5R" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Unity Asset Store to delist all Greater China publishers by 2026</strong>: The move stems from a codebase schism between the global engine and the China-specific Tanji fork, effectively cutting off global access for thousands of assets. <a href="https://app.daily.dev/posts/HcJsvKbOc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PostgreSQL JSONB vs MongoDB for denormalized updates</strong>: Benchmarks show MongoDB’s updateMany is significantly more efficient for write-heavy DDD aggregates, as PostgreSQL JSONB requires rebuilding entire documents and WAL records for every change. <a href="https://app.daily.dev/feed-by-ids?id=IUiWRj9Tc&amp;id=MMhxlqRh1" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Cognition Lab claims the death of the human DevOps engineer</strong>: Cognition Lab is back with a new version of Devin, claiming that human DevOps engineers are now the ‘exception rather than the default.’ It’s a bold take from a company whose primary product is an AI agent, especially when their ‘proof’ is a benchmark they likely tuned for. Expecting a CLI tool to replace entire infrastructure teams is the kind of peak-hype fantasy that ignores the messy reality of legacy systems and human accountability. <a href="https://app.daily.dev/posts/LfBawqtwk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI startups use two-tier valuations to fake unicorn status</strong>: Venture capitalists are now helping startups like Aaru and Serval maintain ‘headline’ valuations of $1B while lead investors actually buy in at less than half that price. This accounting gymnastics allows founders to brag about unicorn status for recruiting while setting themselves up for catastrophic down-rounds if they don’t hit impossible growth targets. It’s a classic bubble behavior where the optics of success are more important than the actual math of the business. <a href="https://app.daily.dev/posts/cRCbCFvQ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The information war of selective AI safety and military deals</strong>: OpenAI is touting a deal for classified military networks while Anthropic plays the ‘ethical’ card by refusing the Pentagon, despite both companies leaking selective financial projections to keep the hype train moving. Anthropic’s safety stance looks increasingly performative when they support other military uses, and OpenAI’s ‘all lawful use’ clause is essentially a blank check for mass surveillance. It’s less about safety and more about which billionaire gets the biggest government contract. <a href="https://app.daily.dev/feed-by-ids?id=CiUt1uJYH&amp;id=egxeYRWKg" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
