Anthony Sistilli

Google Cloud

Hacker News

InfoQ

matklad

NDC Conferences

<h2>Must know</h2>
<ul>
<li><strong>Kubernetes is now the standard for enterprise AI, but culture is the new bottleneck</strong>: Kubernetes production usage has hit 82% among surveyed organizations, solidifying its role as the standard infrastructure for enterprise AI workloads, with 66% of AI adopters using it for inference. The 2025 CNCF Annual Cloud Native Survey, widely cited, reveals that cultural resistance, not technical complexity, is now the primary barrier to adoption. This shift highlights the need for organizational change management alongside technical implementation, as noted by the CNCF. <a href="https://app.daily.dev/posts/Lwsgn3lZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-driven exploit and malware development is accelerating</strong>: Large language models (LLMs) like Opus 4.5 and GPT-5.2 are now generating sophisticated exploits for zero-day vulnerabilities, bypassing modern security mitigations. One experiment showed GPT-5.2 solving a complex scenario in 3 hours for about $50, shifting the bottleneck from human expertise to LLM token throughput. Check Point Research also identified VoidLink, the first known AI-generated sophisticated malware suite, developed in under a week by a single person using AI agents, a task estimated to take 30 weeks for multiple human teams. This indicates AI is industrializing offensive cybersecurity. <a href="https://app.daily.dev/feed-by-ids?id=WkT9zuzY1&amp;id=nLTC7GTyu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Git MCP server had critical vulnerabilities</strong>: Anthropic’s Git MCP server was found to have three critical vulnerabilities, including path validation bypass and argument injection, which could lead to remote code execution via prompt injection. These flaws, discovered in June 2025 and patched by December, highlight new attack surfaces in AI-integrated systems where chaining vulnerabilities with other components can allow attackers to read, delete, or write arbitrary files. This incident underscores the unexpected security risks when AI agent tools involve code execution or file access. <a href="https://app.daily.dev/posts/U6A8HrCeq" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Salesforce migrates 1,000+ EKS clusters to Karpenter</strong>: Salesforce moved over 1,000 EKS clusters from Kubernetes Cluster Autoscaler to Karpenter, cutting scaling latency from minutes to seconds and reducing operational overhead by 80%, leading to improved node utilization and 5% cost savings. <a href="https://app.daily.dev/posts/HWPLkODi3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New AI coding agent GLM-4.7 Flash outperforms competitors</strong>: GLM-4.7 Flash, a 30B-A3B mixture-of-experts language model, significantly outperforms competitors like Qwen 3-30B-A3B in coding and agentic AI benchmarks, boasting superior tool calling abilities and fast inference speeds. <a href="https://app.daily.dev/posts/eWdzokgG0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>EU launches decentralized vulnerability database</strong>: The GCVE has launched <a href="http://db.gcve.eu" target="_blank" rel="noopener nofollow">db.gcve.eu</a>, a public, decentralized cybersecurity vulnerability database integrating data from over 25 sources, aiming to reduce European reliance on US-based systems like CVE. <a href="https://app.daily.dev/posts/i2uCJ3RGE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents for coding: Achievements, pitfalls, and guardrails</strong>: AI coding agents like Claude Code can deliver substantial velocity gains for well-defined tasks, but struggle with ambiguous requirements, often requiring human oversight, strong test coverage, and explicit architectural documentation to avoid technical debt. <a href="https://app.daily.dev/feed-by-ids?id=EqMZfdLoq&amp;id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered observability for OpenShift</strong>: Red Hat released an AI-powered observability tool that translates complex infrastructure metrics into conversational insights for OpenShift AI environments, allowing natural language queries about GPU performance, model throughput, and cluster health. <a href="https://app.daily.dev/posts/jpKDLQQHM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New Agent2Agent protocol and AI agent architecture advances</strong>: The A2A Java SDK 1.0.0.Alpha1 has been released, aligning with the forthcoming Agent2Agent Protocol 1.0, standardizing communication between AI agents and introducing modular, reusable ‘Agent Skills’ to address prompt bloat and scalability issues. <a href="https://app.daily.dev/feed-by-ids?id=LmGT7Q5vx&amp;id=PMa93QJZs&amp;id=ZcWzztsx0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft Intune mandates security updates for mobile apps</strong>: Microsoft will enforce mandatory Intune Mobile Application Management security updates from January 19, 2026, requiring iOS and Android apps to update to the latest Intune SDK versions or be blocked from accessing corporate resources. <a href="https://app.daily.dev/posts/PNEoUht10" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Vibecoding: The next big thing, or just more AI hype?</strong>: Emergent, an Indian AI startup, just bagged $70 million for ‘vibe-coding,’ claiming 5 million users and $50 million ARR. Apparently, this lets executives build full-stack apps without ‘advanced coding expertise.’ An opinion piece even suggests every exec should be ‘vibecoding’ to become an individual contributor. Because nothing says ‘robust software’ like a CEO ‘vibecoding’ an internal tool, right? <a href="https://app.daily.dev/feed-by-ids?id=AllrxKqHW&amp;id=McILqauKu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated code: The gift that keeps on giving… memory leaks</strong>: Turns out, AI-generated SwiftUI code is a master of creating persistent memory leaks by strongly capturing ‘self’ in deeply nested closures. Who knew that AI, in its infinite wisdom, would forget basic memory management best practices? Engineers then get to spend their precious time refactoring, adding weak references, and ownership annotations to stabilize the app. So much for AI making our lives easier, unless ‘easier’ means ‘more debugging.’ <a href="https://app.daily.dev/feed-by-ids?id=0Dxh8ahUn&amp;id=NOlmy6yd2&amp;id=yTmR3XbQ9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI security: Just apply Zero Trust and pray?</strong>: Security researchers found Google Gemini vulnerable to prompt injection via malicious calendar invites, leading to exfiltration of private meeting data. The solution? Apply Zero Trust principles and enforce least privilege on AI systems. So, we’re building AI that can be tricked by a calendar invite, and our best defense is to treat it like a rogue employee. Sounds like a solid plan for the future of enterprise security. <a href="https://app.daily.dev/feed-by-ids?id=dHjNBMkFO&amp;id=5ZpMYMEZm&amp;id=dRAo2Mdf1&amp;id=j6bBmLvCW" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
