<h2>Must know</h2>
<ul>
<li><strong>AI coding assistants demand more upfront planning, not less</strong>: Despite promises of faster code generation, AI coding assistants are actually increasing development times and architectural planning needs. A 2025 study found developers using AI tools took 19% longer to complete tasks, with code review times increasing by 91%. The issue stems from AI’s lack of contextual memory and tendency to produce security vulnerabilities, requiring detailed specifications and human oversight, as reported by The Developing Dev. It seems AI is more like a junior developer that needs constant hand-holding than a self-sufficient coding guru. <a href="https://app.daily.dev/posts/pYjmmMIkn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LLM inference hardware is hitting a memory wall</strong>: The biggest bottleneck for large language model (LLM) inference isn’t compute power, it’s memory capacity and bandwidth, according to a highly reliable preprint. The autoregressive decode phase of LLMs constantly accesses massive model weights, straining memory systems. Future datacenter AI accelerators will need innovations like High Bandwidth Flash, Processing-Near-Memory architectures, and 3D memory-logic stacking to keep up, as current hardware is simply not designed for this kind of memory-bound workload. <a href="https://app.daily.dev/posts/t0LNVAkom" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Netflix is going full TikTok and eyeing Warner Bros Discovery</strong>: Netflix is redesigning its mobile app to feature vertical, short-form video content, directly competing with TikTok and Instagram Reels. This move, launching in late 2026, is a clear response to changing viewer habits. Beyond short videos, Netflix is also expanding into gaming and live sports, and is reportedly pursuing an $82.7 billion acquisition of Warner Bros Discovery to bolster its content library and theatrical distribution. It seems Netflix wants to be everything to everyone, or at least, everything to everyone with a short attention span. <a href="https://app.daily.dev/posts/4L22WBSq0" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Uber uses AI to predict airport rideshare demand</strong>: Uber has rolled out three new forecasting models—ETR, EPH, and Driver Deficit Forecasting—to predict driver queue wait times, earning potential, and supply shortages at airports, using real-time data and Apache Flink streaming pipelines. <a href="https://app.daily.dev/posts/y5uOUdPPc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux kernel gets a continuity plan for Linus Torvalds’ repo</strong>: The Linux 6.19 kernel now includes an official continuity document outlining procedures for managing the mainline Git repository if Linus Torvalds’ becomes unavailable, ensuring project stability and succession planning. <a href="https://app.daily.dev/feed-by-ids?id=EogQGzxtg&amp;id=mNRQJiXzS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Turbopuffer ANN v3 achieves low-latency vector search at scale</strong>: Turbopuffer’s ANN v3 now delivers 200ms p99 query latency with 1k QPS over 100 billion vectors, utilizing hierarchical clustering, binary quantization (RaBitQ), and NVMe SSDs for efficient, scalable vector search. <a href="https://app.daily.dev/posts/EHAeemnne" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PHPStan 2.1.34 boosts performance by up to 50%</strong>: PHPStan 2.1.34 significantly improves performance, with analysis times 25-40% faster in real-world projects and up to 50% faster internally, though it now requires more memory. <a href="https://app.daily.dev/posts/hEVb27hAV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rust-based Homebrew alternative claims 5-20x speedup</strong>: ZeroBrew, an experimental Rust-based package manager, claims to be 5-20 times faster than Homebrew for macOS, offering features like content-addressable storage, parallel downloads, and compatibility with Homebrew’s CDN. <a href="https://app.daily.dev/posts/Mu2gAc20u" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Wayfair uses PostgreSQL as a dead letter queue</strong>: Wayfair implemented PostgreSQL as a Dead Letter Queue (DLQ) for failed Kafka events, storing metadata and using <code>FOR UPDATE SKIP LOCKED</code> for safe, distributed retries, improving visibility and operational simplicity over Kafka-native DLQs. <a href="https://app.daily.dev/posts/eKm9n1imv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>DoorDash optimizes A/B testing with multi-armed bandits</strong>: DoorDash is using multi-armed bandits (MAB) and Thompson sampling to dynamically allocate user traffic to better-performing experiment variants, aiming for faster iteration and reduced opportunity costs in A/B testing. <a href="https://app.daily.dev/posts/FLQQB5Gyi" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI agents will automate your compliance, because nothing says ‘trustworthy’ like an AI</strong>: Apparently, ‘Agentic AI’ is here to save enterprises from the horrors of compliance and ESG reporting. According to a flurry of identical articles, this tech will turn compliance from a ‘cost center into a strategic business advantage’ by automating everything. No mention of who’s actually using it, or if it’s been proven to work, but hey, it sounds good on a slide deck. What could possibly go wrong when you hand over your regulatory obligations to an autonomous AI? <a href="https://app.daily.dev/feed-by-ids?id=Skhkgwj1Z&amp;id=qF5kIahRI&amp;id=uTj3yEY4b" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ChatGPT and Claude are citing ‘Grokipedia,’ because who needs facts when you have AI-generated ‘truth’?</strong>: In a stunning display of AI’s discerning taste, both ChatGPT’s GPT-5.2 and Anthropic’s Claude are reportedly citing ‘Grokipedia’—Elon Musk’s ‘conservative-leaning’ AI-generated encyclopedia—as a source for obscure topics. Critics say Grokipedia is full of inaccuracies and bias, but why let pesky facts get in the way of a good AI-generated answer? It seems our advanced LLMs are just as susceptible to questionable sources as a college student cramming for a midterm. <a href="https://app.daily.dev/posts/SDpas3mop" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Encrypted quantum cloning protocol announced, because defying physics is the new black</strong>: A new quantum protocol is supposedly enabling ‘encrypted cloning of quantum states,’ effectively circumventing the no-cloning theorem. This ‘breakthrough’ promises to revolutionize quantum data storage and security. Conveniently, the article provides zero technical details, no names of researchers, and no mention of experimental verification. It’s almost as if someone just made a bold claim without any pesky evidence to back it up. Quantum physics is hard, but apparently, marketing hype is easy. <a href="https://app.daily.dev/posts/z7OYuDS7t" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
