<h2>Must know</h2>
<ul>
<li><strong>AI coding agents demand more upfront planning and energy</strong>: AI coding assistants, while seemingly accelerating code generation, are creating a productivity paradox. A 2025 study found developers using AI tools took 19% longer to complete tasks, with code review times increasing by 91% due to issues like 45% of AI-generated code containing security vulnerabilities. This necessitates more detailed architectural planning and treating AI like a junior developer requiring thorough oversight, as reported by Tech Lead Digest. Furthermore, these tools are energy hogs: a median Claude Code session consumes about 41 watt-hours, 138 times more than a typical chatbot query, with daily usage comparable to running a dishwasher. <a href="https://app.daily.dev/feed-by-ids?id=pYjmmMIkn&amp;id=ex1Cm3GIg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux kernel gets continuity plan and hardware support</strong>: The Linux 6.19 kernel now includes an official continuity plan for Linus Torvalds’ mainline Git repository, ensuring project resilience with a 72-hour response window for succession procedures, as discussed at the 2025 Linux Maintainer Summit. In parallel, Linux 7.0 is gaining display support for Qualcomm Snapdragon 8 Elite Gen 5 and older Adreno 225 GPUs, alongside AMD graphics driver fixes for various hardware generations, including RDNA4, improving overall hardware compatibility and stability. <a href="https://app.daily.dev/feed-by-ids?id=LVxtNaVxq&amp;id=cxhDtp9rh&amp;id=EogQGzxtg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Turbopuffer ANN v3 achieves massive scale vector search</strong>: Turbopuffer’s ANN v3 now delivers 200ms p99 query latency at 1,000 queries per second over 100 billion vectors (200TiB of data). This is achieved through hierarchical clustering, RaBitQ binary quantization for 16-32x compression, and distributing data across storage-dense VMs utilizing AVX-512 instructions. The system maintains 92% recall by selectively reranking less than 1% of candidates, making large-scale vector search viable for production environments. <a href="https://app.daily.dev/posts/EHAeemnne" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>LLM inference hardware bottlenecks shift to memory</strong>: LLM inference is increasingly bottlenecked by memory capacity and bandwidth during the autoregressive Decode phase, rather than compute resources, pushing research towards High Bandwidth Flash, Processing-Near-Memory, and 3D memory-logic stacking. <a href="https://app.daily.dev/posts/t0LNVAkom" 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 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 DLQ topics. <a href="https://app.daily.dev/posts/eKm9n1imv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Uncloud Kubernetes alternative gains traction</strong>: Uncloud, a Kubernetes alternative built with Docker, Caddy, and WireGuard, is gaining attention among small teams for single-command deployment from source to production, bypassing container registries and aiming for a managed control center in 2026. <a href="https://app.daily.dev/posts/FrS1fa85P" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>MIT develops scalable superconducting nanowire memory</strong>: MIT researchers have developed a 4x4 superconducting nanowire memory array operating at 1.3 Kelvin, achieving a bit error rate of 10⁻⁵ and a functional density of 2.6 megabits per square centimeter, addressing key scalability and error rate challenges for quantum and low-energy computing. <a href="https://app.daily.dev/posts/UFLhnkgMo" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Measurement-free fault-tolerant quantum computing demonstrated</strong>: Researchers have demonstrated a measurement-free approach to fault-tolerant quantum computing using trapped ions, bypassing slow and error-prone mid-circuit measurements to achieve universal logical operations and execute Grover’s algorithm on three logical qubits. <a href="https://app.daily.dev/posts/b76Xs6rDw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Raspberry Pi OS adds SSH over USB for headless setup</strong>: Raspberry Pi OS now includes <code>rpi-usb-gadget</code>, enabling SSH over USB for headless devices, simplifying initial setup by allowing direct connection to a host machine as a network adapter. <a href="https://app.daily.dev/posts/2xZnVNVTE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Multitasking guidance for software engineers</strong>: Software engineers can improve productivity by managing context switching, adopting CPU-like mental practices with ‘safe points,’ batching similar work, and maintaining 2-3 active workstreams, rather than avoiding multitasking entirely. <a href="https://app.daily.dev/posts/3iFQMXtvc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Agentic AI will solve all your compliance woes</strong>: Apparently, ‘Agentic AI’ is the new magic bullet for enterprise compliance and ESG reporting, promising to transform manual, fragmented processes into automated, unified, and auditable workflows. It’s positioned as turning compliance from a cost center into a strategic advantage, minimizing risk and strengthening governance. The articles, all published on the same day, are suspiciously consistent in their positive framing, with no mention of downsides, limitations, or actual companies using this miracle tech. Sounds like a coordinated marketing push for a solution that’s long on promise and short on proof. <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>Quantum cloning protocol defies physics, probably</strong>: A new quantum protocol is reported to enable ‘encrypted cloning’ of quantum states, supposedly circumventing the no-cloning theorem—a foundational principle in quantum information theory. The article, from a single, unnamed source, offers no technical details, no researchers, no institutions, and no experimental evidence. It just makes broad claims about breakthroughs in quantum data storage and security. Color me skeptical; ‘defying’ established physics usually requires a bit more than a vague announcement. <a href="https://app.daily.dev/posts/z7OYuDS7t" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated encyclopedia is totally unbiased, trust us</strong>: ChatGPT’s GPT-5.2 and Anthropic’s Claude are reportedly citing ‘Grokipedia,’ an AI-generated encyclopedia by Elon Musk, especially for obscure topics. Grokipedia is described as ‘conservative-leaning’ and has faced criticism for ‘inaccurate and biased content.’ So, our leading AI models are pulling facts from an AI-generated, potentially biased source, and nobody seems to be questioning the vetting process. What could possibly go wrong when our AI overlords are trained on their own echo chambers? <a href="https://app.daily.dev/posts/SDpas3mop" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
