<h2>Must know</h2>
<ul>
<li><strong>Meta goes big on AMD for AI infrastructure</strong>: Meta has inked a massive multiyear deal with AMD, reportedly worth up to $100 billion, to acquire 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 also 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. This 3B-parameter multimodal model aims for efficient inference directly on mobile, web, and desktop platforms, prioritizing low latency, privacy, and offline capability. It contrasts with existing GUI agents that rely on larger cloud-based models, potentially reducing Siri’s dependence on Google Cloud. <a href="https://app.daily.dev/posts/qj9GoBvQx" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New benchmark for OpenTelemetry and AI agent performance</strong>: Quesma released OTelBench, an open-source benchmarking suite for evaluating OpenTelemetry (OTel) pipeline performance and AI agent effectiveness in Site Reliability Engineering (SRE) tasks. The tool simulates high-load traffic and tests LLMs on complex observability tasks, revealing that LLMs often achieve success rates below 30% in production-grade instrumentation scenarios. This highlights the current limitations of AI in real-world SRE. <a href="https://app.daily.dev/posts/OB1MlP8PG" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<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. <a href="https://app.daily.dev/posts/ZbbWKhN8A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NPM supply chain attack targets AI coding tools</strong>: A new npm supply chain attack, SANDWORM_MODE, targets developer utilities and AI coding tools like Claude Code, harvesting credentials and injecting malicious code into CI pipelines. <a href="https://app.daily.dev/posts/q6c2RLQKH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Hetzner announces significant price hikes</strong>: Hetzner, a major European cloud provider, is increasing prices by up to 50% starting April 1, 2026, due to dramatic increases in hardware costs driven by AI industry demand. <a href="https://app.daily.dev/posts/van1jyyDC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Formae adds multi-cloud support and AI-assisted development</strong>: Platform Engineering Labs updated formae, its open-source IaC platform, with multi-cloud support and a redesigned Plugin SDK, simplifying extension with AI-assisted development. <a href="https://app.daily.dev/posts/oAyNaPkCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitLab extends GPG key expiration</strong>: GitLab extended its GPG key for repository metadata from February 27, 2026, to February 6, 2028, to ensure integrity and minimize user disruption. <a href="https://app.daily.dev/posts/JTFdrzyr9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Node.js 25.7.0 released with new features</strong>: Node.js 25.7.0 was released, featuring HTTP/2 <code>http1Options</code>, ESM entry points for Single Executable Applications, and a release candidate for the built-in SQLite module. <a href="https://app.daily.dev/posts/K4geCuziz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent browser sandbox project announced</strong>: A new project named “browser sandbox” was announced, aiming to create a secure browser environment for AI agents to interact with the web with “zero config” setup. <a href="https://app.daily.dev/posts/AY0JUdVv6" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Pentagon-Anthropic AI standoff over military use</strong>: The Pentagon issued a Friday deadline to Anthropic to remove AI guardrails on Claude for unrestricted military use, threatening to designate them a ‘supply chain risk’ if they refuse. Apparently, the DOD is so dependent on Anthropic for ‘classified frontier AI access’ that they’re willing to throw around threats usually reserved for foreign adversaries. Because nothing says ‘innovation’ like strong-arming your vendors into compromising their ethical stances. <a href="https://app.daily.dev/posts/vbdfUwGoN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft and Salesforce confirm ‘industry suspicions’ about AI</strong>: Microsoft Research and Salesforce AI reportedly analyzed over 200,000 AI conversations, confirming ‘industry suspicions’ but providing absolutely no details on what those suspicions were or what was found. This vague announcement, sourced from a retweet, is a masterclass in generating curiosity without delivering any actual information. It’s like saying ‘we found something interesting!’ and then walking away. <a href="https://app.daily.dev/posts/FAzFsVs8B" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI models in Cursor get ‘up to 30% faster’</strong>: OpenAI models integrated into Cursor were upgraded, reportedly resulting in ‘up to 30% faster performance’ for all users. This claim, originating from a promotional tweet, lacks any independent verification. So, while it sounds great, it’s probably best to take that ‘up to 30%’ with a grain of salt, or perhaps a whole shaker. <a href="https://app.daily.dev/posts/CNQCNZ2N0" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
