<h2>Must know</h2>
<ul>
<li><strong>Apple opens up the Neural Engine for local inference</strong>: Apple is doubling down on its 8-year hardware strategy by exposing its built-in language models via the FoundationModels framework in macOS 26. A new open-source CLI tool called apfel now allows developers to run local inference on the Mac’s Neural Engine through an OpenAI-compatible HTTP server, bypassing the need for cloud subscriptions or API keys. This vertical integration from silicon to Core ML provides a privacy-first alternative to cloud-only AI approaches, as reported by various technical deep dives. <a href="https://app.daily.dev/feed-by-ids?id=VuypgGktt&amp;id=YklInVgqT" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub slashes pull request memory usage by half</strong>: The GitHub engineering team successfully refactored their React-based diff architecture to handle massive 10,000-line pull requests without crashing the browser. By reducing components from 8 to 2 per line and implementing TanStack Virtual for windowing, they achieved a 74% reduction in rendered components and a 78% improvement in Interaction to Next Paint (INP). The GitHub Blog details how moving complex state to conditionally rendered children and adopting O(1) Map lookups solved their 1 GB JavaScript heap issues. <a href="https://app.daily.dev/posts/DMKgp6Vib" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Coupang scales DynamoDB sequences to 50k per second</strong>: To migrate 100+ services from relational databases to DynamoDB, Coupang engineered a two-tier caching architecture that serves 99% of sequence requests from local application memory. This system uses a sliding window algorithm to predict cache refills, resulting in only 10-20 DynamoDB writes per second despite the massive throughput. InfoQ notes that the project succeeded by challenging the assumption that all services required gap-free or globally ordered sequences, collapsing a complex coordination problem into a simple SDK. <a href="https://app.daily.dev/posts/TEv7fV07V" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Kilo Code launches production ready VS Code extension</strong>: The rebuilt extension features parallel tool calls, sub-agent delegation, and Git worktree isolation for parallel agent tasks. <a href="https://app.daily.dev/posts/i3oAoFhPq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Meta patches Linux kernel to prevent TCP throttling</strong>: A new order-aware vmpressure patch prevents high-order memory fragmentation from incorrectly signaling memory pressure and slowing network throughput. <a href="https://app.daily.dev/posts/cG7lwrIeI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>WordPress introduces esbuild based plugin build tooling</strong>: The new @wordpress/build tool replaces webpack and Babel with zero-config convention-based discovery and auto-generated PHP script registration. <a href="https://app.daily.dev/posts/KdxAIboWK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vulkan 1.4.348 adds TOSA tensor operator support</strong>: The update includes new extensions for neural network data graphs and custom primitive restart indices to improve OpenGL emulation. <a href="https://app.daily.dev/posts/ZChBOyJvY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Convex launches enterprise tier with cheaper data egress</strong>: The platform is shifting from file bandwidth pricing to a $0.11/GB egress metric while increasing concurrent deployment limits for all users. <a href="https://app.daily.dev/posts/a32x0VNOX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Laravel Activitylog v5 requires PHP 8.4</strong>: The update introduces a unified trait for logging, activity buffering for bulk inserts, and a dedicated column for attribute changes. <a href="https://app.daily.dev/posts/vYESRvPGN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Periodic Labs applies AI foundation models to materials science</strong>: Former OpenAI VP Liam Fedus is building a closed-loop system that combines physics simulations with LLMs to solve atomic-scale engineering bottlenecks. <a href="https://app.daily.dev/posts/UOMOIFzJc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The death of the AI wrapper</strong>: Alex CloudStar claims ‘AI wrappers are dead’ because big providers are absorbing their features, yet the proposed solution is ‘Vertical AI’—which is usually just a wrapper with a more expensive domain name. While the 90-day churn stats are likely real, pretending that ‘context engineering’ is a magical moat rather than just a fancy term for better RAG is peak industry cope. <a href="https://app.daily.dev/posts/JEAdPfYqt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>You are a factory worker not a developer</strong>: The Serious CTO is back with a lecture on how writing code is a ‘liability’ and you’re basically a blue-collar worker if you don’t estimate maintenance costs for every line. It’s the classic ‘stop building things’ philosophy that sounds great in a blog post but results in six months of meetings to decide on a single database schema change. <a href="https://app.daily.dev/posts/z80At6eoi" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Agentic commerce is the future of retail</strong>: Stripe and Shoptalk are pushing ‘agentic commerce’ as the next big thing, where AI agents will apparently handle your checkout and product discovery. In reality, it’s a desperate attempt to make chatbots sound like a revenue stream by adding a ‘Buy Now’ button to a hallucinating LLM that can’t even tell if a shirt is actually in stock. <a href="https://app.daily.dev/posts/dnqTlZJ1u" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
