<h2>Must know</h2>
<ul>
<li><strong>Hetzner hikes prices up to 50% due to AI-driven hardware costs</strong>: Hetzner, a major European budget cloud provider, is increasing prices by up to 50% starting April 1, 2026, across all regions. This significant hike is attributed to dramatic increases in hardware costs, particularly for DRAM, NAND flash, and server CPUs, driven by the AI industry’s demand. Unlike hyperscalers with long-term vendor contracts, smaller providers like Hetzner are directly passing these costs to users, with RAM prices reportedly rising 80-400%. <a href="https://app.daily.dev/posts/van1jyyDC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Meta inks massive AI chip deal with AMD</strong>: Meta has signed a multiyear deal with AMD, reportedly worth up to $100 billion, to expand its data center infrastructure with AMD’s MI540 GPUs and CPUs. This move aims to add 6GW of compute capacity for AI model development, diversifying Meta’s AI compute stack beyond Nvidia. The agreement includes a performance-based warrant for AMD to issue Meta up to 160 million shares, contingent on AMD’s stock reaching $600. <a href="https://app.daily.dev/posts/vYZNsZv8f" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Apple unveils on-device UI AI model</strong>: Apple researchers introduced Ferret-UI Lite, a lightweight, on-device AI model designed to understand and interact with user interfaces without cloud processing. This 3B-parameter multimodal model aims for low latency, privacy, and offline capability across mobile, web, and desktop platforms. It achieved 91.6% on the ScreenSpot-V2 GUI grounding benchmark, competing with much larger models, though it struggles with long-horizon multi-step tasks. <a href="https://app.daily.dev/posts/qj9GoBvQx" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Debian: the reliable distro</strong>: Debian is highlighted as a stable, reliable Linux distribution with a large repository and community, focusing on reliability over cutting-edge features. <a href="https://app.daily.dev/posts/2R3XlQZuI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rust TUI for PCAP analysis</strong>: A new Rust-built TUI tool called Babyshark offers a flows-first approach to analyzing PCAP network captures, providing features like live capture, anomaly detection, and plain-English flow explanations. <a href="https://app.daily.dev/posts/v12mxt3cf" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cilium 1.19 boosts Kubernetes networking</strong>: Cilium 1.19 enhances Kubernetes networking with BPF Host Routing for IPsec, Gateway API v1.4 support, stable Multi-Pool IPAM, and initial Ztunnel integration for ambient mesh-style encryption. <a href="https://app.daily.dev/posts/zYLvcHpYh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Node.js 25.7.0 released</strong>: Node.js 25.7.0 introduces HTTP/2 <code>http1Options</code>, ESM support for Single Executable Applications, and an upgraded npm to 11.10.1. <a href="https://app.daily.dev/posts/K4geCuziz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Elasticsearch v9 upgrades vector search performance</strong>: Elasticsearch v9.2.1 shows a 4x performance boost for vector search compared to v8.18.3, with significantly higher throughput and lower latency under high concurrency. <a href="https://app.daily.dev/posts/Iw8wLbMFe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Coreboot ported to ThinkPad X270</strong>: A developer successfully ported Coreboot/Libreboot to the ThinkPad X270, including neutralizing the Intel Management Engine and resolving PCIe issues for NVMe and WiFi. <a href="https://app.daily.dev/posts/4u0sx2QoJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>DMARCbis update aims for clearer email authentication</strong>: DMARCbis, an upcoming update to the DMARC email authentication standard, introduces clearer reporting expectations and standardized DNS tree walks for receiver-side policy discovery. <a href="https://app.daily.dev/posts/yEFoBF8zN" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Google’s ‘vibe-coding’ app gets ‘agentic’ AI</strong>: Google upgraded its ‘vibe-coding app’ Opal with new agentic AI workflows, allowing users to create automated workflows using text prompts powered by Gemini 3 Flash. Apparently, this ‘Agent step’ analyzes goals and calls tools automatically, which sounds suspiciously like every other ‘AI agent’ demo that falls apart when faced with reality. But hey, ‘vibe-coding’! <a href="https://app.daily.dev/posts/ZbbWKhN8A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft and Salesforce confirm ‘AI suspicions’ with zero details</strong>: Microsoft Research and Salesforce AI supposedly analyzed over 200,000 AI conversations and ‘confirmed industry suspicions.’ What suspicions? What findings? Who knows! The social media post from @rryssf_ is so vague it’s practically a masterclass in generating curiosity without delivering any actual information. Perhaps the suspicion confirmed is that AI conversations are mostly hot air. <a href="https://app.daily.dev/posts/FAzFsVs8B" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents fail at SRE tasks, study finds</strong>: Quesma released OTelBench, a benchmarking suite that evaluates OpenTelemetry pipelines and, more importantly, AI agent effectiveness in SRE tasks. Their findings indicate LLMs often achieve success rates below 30% in production-grade instrumentation scenarios. So, while everyone’s hyping AI for ops, the actual data suggests it’s mostly just generating more problems than solutions. Shocking. <a href="https://app.daily.dev/posts/OB1MlP8PG" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
