<h2>Must know</h2>
<ul>
<li><strong>Meta goes big on AMD for AI infrastructure</strong>: Meta just inked a massive multiyear deal with AMD, reportedly worth up to $100 billion, to load its data centers with AMD’s MI540 GPUs and CPUs. This move aims to add 6GW of compute capacity to accelerate AI model development, diversifying Meta’s AI compute stack beyond Nvidia. The agreement even includes a performance-based warrant for AMD to issue Meta up to 160 million shares, contingent on AMD’s stock hitting $600, showing Meta is serious about its ‘personal superintelligence’ goal. <a href="https://app.daily.dev/posts/vYZNsZv8f" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pentagon-Anthropic AI standoff over military use</strong>: The Pentagon is playing hardball with Anthropic, demanding they remove AI guardrails from Claude for unrestricted military use, including mass surveillance and autonomous weapons. Anthropic is refusing, leading the Pentagon to threaten a ‘supply chain risk’ designation or even invoking the Defense Production Act. This standoff highlights the military’s reliance on a single vendor for classified frontier AI and the ethical tightrope these companies walk. <a href="https://app.daily.dev/posts/vbdfUwGoN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Hetzner hikes prices up to 50% due to AI-driven hardware costs</strong>: Hetzner, a major European budget cloud provider, is jacking up prices by as much as 50% starting April 1, 2026, across all its regions. The culprit? Soaring hardware costs, particularly for DRAM, NAND flash, and server CPUs, driven by the insatiable demand from the AI industry. This hits smaller providers harder than hyperscalers like AWS or Google, who are insulated by long-term vendor contracts, forcing Hetzner to pass these costs directly to users. <a href="https://app.daily.dev/posts/van1jyyDC" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Apple unveils on-device UI AI model</strong>: Apple researchers introduced Ferret-UI Lite, a lightweight 3B-parameter multimodal AI model designed for on-device UI understanding and interaction across mobile, web, and desktop, prioritizing low latency and privacy. <a href="https://app.daily.dev/posts/qj9GoBvQx" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google enhances Opal with agentic AI workflows</strong>: Google upgraded its Opal Workflow builder with new agentic capabilities, allowing users to create automated workflows using text prompts, powered by Gemini 3 Flash, and integrated into the Gemini web app. <a href="https://app.daily.dev/posts/ZbbWKhN8A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Quesma releases OTelBench for OpenTelemetry and AI evaluation</strong>: Quesma released OTelBench, an open-source benchmarking suite to evaluate OpenTelemetry pipeline performance and AI agent effectiveness in SRE tasks, revealing LLMs often achieve success rates below 30% in production-grade scenarios. <a href="https://app.daily.dev/posts/OB1MlP8PG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Formae IaC platform expands with multi-cloud and AI-assisted development</strong>: Platform Engineering Labs updated formae, its open-source IaC platform, with multi-cloud support and a redesigned Plugin SDK, enabling AI-assisted development and avoiding fragile state files by continuously codifying infrastructure. <a href="https://app.daily.dev/posts/oAyNaPkCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Node.js 25.7.0 released with SEA and HTTP/2 updates</strong>: Node.js 25.7.0 is out, bringing HTTP/2 <code>http1Options</code>, ESM entry point 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 delivers 4x performance for vector search</strong>: Trendyol achieved a 4x performance boost by upgrading to Elasticsearch v9.2.1 for vector search, showing significantly higher throughput and lower latency compared to v8.18.3 for hybrid search workloads. <a href="https://app.daily.dev/posts/Iw8wLbMFe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>DMARCbis update discussion released</strong>: DMARCbis, an upcoming update to the DMARC email authentication standard, introduces clearer reporting expectations and standardized DNS tree walks to reduce ambiguity, as discussed in a fireside chat by Sendmarc. <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>NPM supply chain attack gets a ‘Shai-Hulud-style’ clickbait title</strong>: Socket researchers ‘uncovered’ an npm supply chain attack, SANDWORM_MODE, involving 19 typosquatted packages impersonating developer tools and ‘AI coding tools’ like Claude Code. The article title, ‘Shai-Hulud-style NPM worm hits CI pipelines and AI coding tools,’ is just pure, unadulterated clickbait. Because nothing says ‘serious security threat’ like referencing a giant sandworm from a sci-fi novel. <a href="https://app.daily.dev/posts/q6c2RLQKH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI models in Cursor get ‘up to 30% faster’ with WebSockets</strong>: OpenAI models integrated into Cursor were ‘upgraded’ to use WebSockets with the Responses API, supposedly resulting in ‘up to 30% faster performance.’ This claim, from a promotional tweet by @leerob, lacks any independent verification. So, your AI coding assistant is faster, maybe? Or maybe it’s just marketing fluff making you feel better about your ‘upgraded’ experience. <a href="https://app.daily.dev/posts/CNQCNZ2N0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Future ‘zero config’ AI agent browser sandbox announced with truncated details</strong>: <code>@nickscamara_</code> announced a new ‘browser sandbox’ project for AI agents to ‘securely’ interact with the web, featuring ‘zero config’ setup. The announcement date is in the future (2026-02-24), and the details are ‘truncated in the source.’ So, we have a future announcement for a vague project with incomplete information, promising ‘zero config’ for AI agents. Sounds like a solid plan, definitely not vaporware. <a href="https://app.daily.dev/posts/AY0JUdVv6" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
