<h2>Must know</h2>
<ul>
<li><strong>Meta goes big on AMD for AI infrastructure</strong>: Meta has signed a multiyear deal with AMD, reportedly worth up to $100 billion, to acquire MI540 GPUs and CPUs. This massive investment aims to add 6GW of compute capacity to accelerate AI model development, diversifying Meta’s AI infrastructure beyond Nvidia and supporting its goal of delivering ‘personal superintelligence.’ <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, 3B-parameter multimodal AI model designed for on-device UI understanding and interaction. This model aims for efficient, low-latency inference across mobile, web, and desktop platforms without cloud reliance, potentially reducing Siri’s dependence on external cloud services. <a href="https://app.daily.dev/posts/qj9GoBvQx" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New tool benchmarks OpenTelemetry and AI agent performance</strong>: Quesma released OTelBench, an open-source benchmarking suite to evaluate OpenTelemetry pipeline performance and AI agent effectiveness in SRE tasks. Findings indicate that LLMs often achieve success rates below 30% in production-grade observability scenarios, highlighting current limitations in AI’s practical application for complex 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>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 that leverages AI-assisted development for rapid extension. <a href="https://app.daily.dev/posts/oAyNaPkCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New technique triples LLM inference speed</strong>: Researchers developed a multi-token prediction technique that reportedly triples LLM inference speed without auxiliary draft models, embedding acceleration directly into model weights. <a href="https://app.daily.dev/posts/hxaAjvILy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Debezium joins GSoC 2026 with AI integration projects</strong>: Debezium is participating in Google Summer of Code 2026, offering projects including a source connector for Milvus and PyDebeziumAI for AI integration via Python. <a href="https://app.daily.dev/posts/XH8DOZHBb" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>JFrog offers 99.99% uptime SLA for software supply chain</strong>: JFrog introduced Premium Availability, offering a 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>Spark 4.1 introduces declarative pipelines</strong>: Apache Spark 4.1 introduces Spark Declarative Pipelines, extending declarative programming to entire data pipelines to reduce boilerplate and automate incremental processing and data quality checks. <a href="https://app.daily.dev/posts/4b5vqOEib" 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 2026 to February 2028 to ensure integrity and minimize disruption for users. <a href="https://app.daily.dev/posts/JTFdrzyr9" 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 deadline to Anthropic to remove AI guardrails on Claude for unrestricted military use, threatening to label them a ‘supply chain risk’ if they refuse. Apparently, the DOD is so dependent on a single vendor for ‘classified frontier AI access’ that they’re willing to strong-arm a company into removing ethical safeguards. Because nothing says ‘responsible AI’ like forcing a vendor to enable mass surveillance and autonomous weapons. <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, targeting developer utilities and AI coding tools like Claude Code. The sensationalist title, ‘Shai-Hulud-style NPM worm hits CI pipelines and AI coding tools,’ makes it sound like a sci-fi epic, but it’s just another supply chain attack harvesting tokens and credentials. The ‘dormant polymorphic engine’ and ‘dead switch home-directory wiper’ are inactive by default, so it’s more of a whimper than a worm, for now. <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</strong>: Microsoft Research and Salesforce AI reportedly analyzed over 200,000 AI conversations, confirming ‘industry suspicions’ about AI. The social media post announcing this provides zero details on what those suspicions are or what was actually found. It’s the perfect clickbait: generate curiosity without delivering any actual information. So, they confirmed something, but we’ll never know what. Riveting. <a href="https://app.daily.dev/posts/FAzFsVs8B" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
