80 LEVEL

Coding with Lewis

databricks

Datadog

Debezium

Hacker News

<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 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 hitting $600, signaling a deep strategic partnership. <a href="https://app.daily.dev/posts/vYZNsZv8f" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Hetzner price hikes signal AI’s hardware cost impact</strong>: Hetzner, a major European cloud provider, is hiking prices by up to 50% starting April 1, 2026, citing dramatic increases in hardware costs for DRAM, NAND flash, and server CPUs. The AI industry’s insatiable demand is straining global supply chains, disproportionately affecting smaller providers like Hetzner and OVHcloud who lack the long-term vendor contracts that insulate hyperscalers like AWS and Google. This means smaller players are passing those costs directly to users. <a href="https://app.daily.dev/posts/van1jyyDC" 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, 3B-parameter multimodal AI model designed for on-device UI understanding and interaction. This model aims for efficient inference directly on mobile, web, and desktop platforms, prioritizing low latency, privacy, and offline capability. It contrasts with larger, cloud-dependent GUI agents and could significantly reduce Siri’s reliance on external cloud services. <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>OTelBench evaluates OpenTelemetry and AI for SRE</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 instrumentation scenarios. <a href="https://app.daily.dev/posts/OB1MlP8PG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google’s Opal gets agentic AI workflows</strong>: Google upgraded its Opal Workflow builder with new agentic capabilities, including an “Agent step” that analyzes goals, determines approaches, and calls tools automatically, powered by Gemini 3 Flash. <a href="https://app.daily.dev/posts/ZbbWKhN8A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New technique triples LLM inference speed</strong>: A new multi-token prediction technique triples LLM inference speed without auxiliary draft models by fine-tuning pretrained models with a student-teacher RL-inspired paradigm and a confidence-adaptive decoding strategy. <a href="https://app.daily.dev/posts/hxaAjvILy" 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 encrypted traffic, Gateway API v1.4 support, stable Multi-Pool IPAM, and initial Ztunnel integration for ambient mesh-style transparent encryption. <a href="https://app.daily.dev/posts/zYLvcHpYh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Spark 4.1 introduces declarative pipelines</strong>: Spark Declarative Pipelines (SDP), introduced in Apache Spark 4.1, extends declarative programming to entire data pipelines, automating incremental processing, dependency ordering, and data quality checks to reduce boilerplate. <a href="https://app.daily.dev/posts/4b5vqOEib" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>JFrog offers 99.99% uptime SLA</strong>: JFrog now offers a Premium Availability tier with a contractual 99.99% in-region uptime SLA for its software supply chain platform, targeting organizations with AI workloads and strict regulatory requirements. <a href="https://app.daily.dev/posts/6JJpZNvHL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Babyshark: a Rust TUI for PCAP analysis</strong>: Babyshark is a new open-source terminal UI (TUI) for analyzing PCAP network captures, built in Rust, offering a flows-first approach with anomaly detectors and plain-English flow explanations. <a href="https://app.daily.dev/posts/v12mxt3cf" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Pentagon demands Anthropic remove AI guardrails</strong>: The Pentagon issued a Friday deadline to Anthropic to remove AI guardrails on Claude for “unrestricted military use,” threatening to label them a “supply chain risk” if they refuse. Because apparently, the DOD is so dependent on a single vendor for classified frontier AI that they’re willing to strong-arm them into removing ethical safeguards. Who needs a backup plan when you can just threaten your suppliers? <a href="https://app.daily.dev/posts/vbdfUwGoN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NPM supply chain attack targets AI coding tools</strong>: Socket researchers uncovered a new npm supply chain attack, SANDWORM_MODE, involving 19 typosquatted packages impersonating developer utilities and AI coding tools like Claude Code. The malware harvests credentials and injects malicious code into CI pipelines, even deploying an MCP server for prompt-injection attacks. Because nothing says “secure development” like your AI coding assistant silently exfiltrating your SSH keys. <a href="https://app.daily.dev/posts/q6c2RLQKH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft and Salesforce confirm AI suspicions (details withheld)</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 is vague, offering zero details, which is a classic move to generate curiosity without actually delivering any information. It’s almost as if they want you to click, but not actually learn anything useful. <a href="https://app.daily.dev/posts/FAzFsVs8B" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
