<h2>Must know</h2>
<ul>
<li><strong>AI productivity gains remain elusive for most businesses</strong>: A survey of nearly 6,000 executives across the US, UK, Germany, and Australia reveals that over 80% of firms have seen no measurable impact on productivity or employment from AI over the past three years. Despite 69% of businesses using some form of AI, the anticipated 1.4% productivity gains and 1.75 million job eliminations by 2028 remain largely theoretical, highlighting a significant gap between AI investment hype and actual business outcomes. This aligns with similar findings from PwC and Deloitte, and a UK government trial of Microsoft M365 Copilot also found no productivity improvement. <a href="https://app.daily.dev/posts/rfmYnNc4e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical vulnerabilities found in popular VS Code extensions</strong>: OX Security disclosed critical and high-severity vulnerabilities in four widely used VS Code extensions: Live Server, Code Runner, Markdown Preview Enhanced, and Microsoft’s Live Preview, collectively impacting 128 million installations. These flaws enable file theft, remote code execution, and local network reconnaissance. While Microsoft silently patched Live Preview, three other maintainers never responded to disclosure notices, leaving millions of developers exposed. This highlights significant supply chain risks in developer tooling. <a href="https://app.daily.dev/posts/IuDICegHN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft Copilot chat security flaw exposes confidential emails</strong>: Microsoft confirmed a security bug (CW1226324) in Copilot Chat that allowed the AI to bypass confidentiality labels and Data Loss Prevention (DLP) policies, summarizing emails from users’ Sent Items and Drafts folders without authorization. The flaw, active since late January, was attributed to an unspecified ‘code issue’ and a fix began rolling out in early February. This incident, reported by BleepingComputer, adds to ongoing enterprise anxiety about AI tools accessing sensitive corporate data, with the European Parliament already blocking similar AI features on work devices. <a href="https://app.daily.dev/posts/jSmIdhAQU" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Databricks launches custom agents for production AI</strong>: Databricks released Custom Agents, formerly Agent Framework, enabling developers to build, test, and deploy production-quality AI agents as fully managed Databricks Apps on serverless compute, with built-in memory and enterprise data access. <a href="https://app.daily.dev/posts/QJn2roYcs" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Cloud launches managed MCP servers for databases</strong>: Google Cloud introduced managed Model Context Protocol (MCP) servers for its major database products, including AlloyDB, Spanner, and Cloud SQL, allowing secure, IAM-authenticated access to operational data for AI agents without infrastructure deployment. <a href="https://app.daily.dev/posts/wLx806hAH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA and Sarvam AI achieve 4x inference speedup for multilingual MoE model</strong>: Sarvam AI, an Indian generative AI startup, collaborated with NVIDIA to optimize inference for its Sovereign 30B multilingual Mixture-of-Experts model, achieving a 4x total speedup on H100 and Blackwell GPUs through kernel-level optimizations and NVFP4 quantization. <a href="https://app.daily.dev/posts/yKP3YkWSF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Red Hat launches enterprise Podman Desktop</strong>: Red Hat released an enterprise-supported build of Podman Desktop, an open-source container development tool, to meet the demand from regulated industries and risk-averse enterprises for vendor-backed support for open-source tooling. <a href="https://app.daily.dev/posts/iEOJ6oeIq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Tailscale integrates with Huntress and Fleet for automated device posture</strong>: Tailscale now offers general availability integrations with Huntress EDR and Fleet MDM, allowing device posture and management state to be used in Tailscale access policies to automate access restrictions for non-compliant devices. <a href="https://app.daily.dev/feed-by-ids?id=L7iJSIIDS&amp;id=VukD3HNtV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Datadog launches sheets for security reporting</strong>: Datadog introduced ‘Datadog Sheets,’ a spreadsheet-style interface for security teams to centralize, filter, and export vulnerability and compliance data, integrating with Code Security and Cloud Security to generate audit-ready reports. <a href="https://app.daily.dev/posts/zptmCv07k" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Apache Flink Kubernetes Operator 1.14.0 released with blue/green deployments</strong>: Apache Flink Kubernetes Operator 1.14.0 introduces native Blue/Green deployment support, enabling zero-downtime upgrades for stateful streaming applications by automatically managing savepoints and traffic switching. <a href="https://app.daily.dev/posts/RfrUoGk5N" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic’s Claude C compiler: A triumph of marketing over reality</strong>: Anthropic proudly announced Claude built a C compiler ‘from scratch’ with ‘no human intervention’ that could compile Linux and Doom. Turns out, it used 37 years of GCC test suites, an online GCC oracle, required human restarts, and cost $20,000 in API calls. Oh, and it couldn’t actually boot Linux. But hey, 16 AI agents running for two weeks to produce 100,000 lines of code is still pretty neat, if you ignore the ‘dishonest marketing language.’ <a href="https://app.daily.dev/posts/0csgEkRMy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Spotify CEO says senior devs don’t code anymore, just supervise AI</strong>: Spotify’s CEO claims senior developers haven’t written code since December 2025, instead supervising AI-generated code. Apparently, AI handles 30% of coding work but only reduces the workforce by 10%, because development is so much more than just typing. Meanwhile, engineers are reporting AI fatigue from reviewing mountains of generated code, often finding it faster to just write it themselves. Who knew AI could be a ‘debt accelerator’ if you don’t have your fundamentals down? <a href="https://app.daily.dev/posts/7hHra6Ai0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated open source pull requests are ‘draining and demoralizing’</strong>: Open-source maintainers, like Godot engine’s Rémi Verschelde, are drowning in ‘draining and demoralizing’ AI-generated pull requests. These LLM-generated contributions are often nonsensical, verbose, and waste reviewer time, leading to an ‘AI slop’ problem. GitHub, partly blamed for incentivizing this behavior, is now scrambling to implement mitigations like PR deletion and automated triage. Because nothing says ‘innovation’ like AI making more work for humans. <a href="https://app.daily.dev/posts/HouoEDJq7" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
