<h2>Must know</h2>
<ul>
<li><strong>Kubernetes is now the standard for enterprise AI, but culture is the new bottleneck</strong>: The 2025 CNCF Annual Cloud Native Survey reveals Kubernetes production usage has hit 82%, solidifying its role as the infrastructure standard for enterprise AI workloads, with 66% of AI adopters using it for inference. However, the survey, widely cited by sources, highlights a critical shift: cultural resistance, not technical complexity, is now the primary barrier to adoption. Mature organizations are distinguishing themselves through GitOps and platform engineering practices, emphasizing that successful AI integration is less about tech and more about people. <a href="https://app.daily.dev/posts/Lwsgn3lZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI is industrializing exploit and malware development</strong>: New research indicates AI models like Opus 4.5 and GPT-5.2 can generate over 40 distinct exploits for zero-day vulnerabilities, even bypassing modern security mitigations like ASLR and CFI. GPT-5.2 solved the hardest scenario in just three hours for about $50, showing exploit generation is now limited by token throughput, not human expertise. Check Point Research also found VoidLink, the first 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 human teams. This confirms AI is rapidly accelerating offensive cyber capabilities. <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 server vulnerabilities exposed RCE risks</strong>: Anthropic’s Git MCP server had three critical vulnerabilities, including path validation bypass and argument injection, which could be chained via prompt injection to achieve remote code execution. These flaws, discovered and patched in 2025, allowed attackers to read, delete, or write arbitrary files. The incident highlights new security risks in AI-integrated systems, especially when chaining AI agent tools that involve code execution or file access, as reported by official sources. <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>North Korean hackers target fintech developers via VS Code projects</strong>: North Korean threat actors are exploiting VS Code project files to deploy backdoors and malware, targeting developers in crypto, blockchain, and fintech by tricking them into cloning infected repositories that automatically execute payloads upon opening. <a href="https://app.daily.dev/posts/1wKLyk8gG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>EU launches decentralized vulnerability database</strong>: The EU has launched <a href="http://db.gcve.eu" target="_blank" rel="noopener nofollow">db.gcve.eu</a>, a public cybersecurity vulnerability database aiming to reduce reliance on US systems like CVE, featuring a decentralized approach with data from over 25 sources and autonomous identifier assignment. <a href="https://app.daily.dev/posts/i2uCJ3RGE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft mandates Intune 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 face immediate blocking from corporate resources. <a href="https://app.daily.dev/posts/PNEoUht10" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding agents show velocity gains but need guardrails</strong>: AI coding agents like Claude Code can deliver 10-100x velocity gains for well-defined tasks, but struggle with ambiguous requirements, often producing repetitive code and requiring significant human oversight and guardrails like automated quality checks and peer reviews. <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>New Python package manager ‘uv’ challenges ‘pip’</strong>: The new Python package manager ‘uv’ offers faster installs, reproducible environments, and better uninstall mechanisms compared to ‘pip’, prompting developers to evaluate based on project needs and workflow efficiency. <a href="https://app.daily.dev/posts/Y2d0smXfY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PHP 8.4 changes break <code>http_build_query()</code> behavior</strong>: A PHP 8.4 update unexpectedly changed <code>http_build_query()</code> behavior, causing runtime errors with non-backed enums and leading to community frustration over lack of communication, prompting the release of a URI toolkit library with compatibility modes. <a href="https://app.daily.dev/posts/FEB00NeSK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Reticulum offers cryptography-based networking for unstoppable networks</strong>: Reticulum is an open-source, cryptography-based networking stack that enables decentralized networks over diverse physical mediums like LoRa and WiFi, providing end-to-end encryption and multi-hop routing with extremely low bandwidth requirements, operating independently of IP networks. <a href="https://app.daily.dev/posts/TzmFHVP5g" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI-powered ‘vibecoding’ startup Emergent raises $70M, because we needed more buzzwords</strong>: Emergent, an Indian AI-powered ‘vibecoding’ startup, just bagged $70 million in Series B funding, valuing it at $300 million, claiming $50 million ARR and 5 million users. Apparently, ‘vibecoding’ lets you build full-stack apps with AI agents, reducing the need for actual coding expertise. An opinion piece even suggests executives should be ‘vibecoding’ to build internal tools, because who needs engineers when you have AI and a good ‘vibe’? <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 SwiftUI code: because who needs best practices?</strong>: Turns out, AI-generated SwiftUI code is a masterclass in creating memory leaks by strongly capturing <code>self</code> in deeply nested closures, leading to unbounded memory growth and production issues. Engineers had to manually refactor with weak references and explicit capture lists, proving that while AI can churn out code, it still needs a human to teach it basic memory management and prevent it from turning your app into a memory hog. <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>Google Gemini prompt injection: because AI security is just a suggestion</strong>: Security researchers found a vulnerability in Google Gemini where malicious calendar invitations could exploit the model’s automatic ingestion of event data, allowing attackers to manipulate its behavior and exfiltrate private meeting data. This highlights the ‘unique’ challenge of prompt injection in LLMs, where contextual meaning, not traditional code exploits, is the attack vector. So, while Google ‘mitigated’ this specific flaw, it’s a stark reminder that AI systems are still basically trusting everything you throw at them. <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>
