<h2>Must know</h2>
<ul>
<li><strong>Kubernetes is now the standard for enterprise AI, 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. Interestingly, the survey highlights that cultural resistance, not technical complexity, is now the primary barrier to adoption. This shift underscores that while the tech is mature, organizational dynamics are lagging, as reported by the CNCF. <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>: Experiments show LLMs like Opus 4.5 and GPT-5.2 can generate over 40 distinct exploits for zero-day vulnerabilities, bypassing modern security mitigations in hours for minimal cost. Check Point Research also identified VoidLink, the first AI-generated sophisticated malware suite, developed in a week by a single person using AI agents. This indicates AI is rapidly accelerating offensive cyber capabilities, shifting the bottleneck from human expertise to LLM token throughput, as reported by Hacker News. <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 had critical RCE vulnerabilities</strong>: Anthropic’s Git MCP server was found to have three critical vulnerabilities (CVE-2025-68145, CVE-2025-68144, CVE-2025-68143) that could be chained via prompt injection to achieve remote code execution. These flaws allowed attackers to read, delete, or write arbitrary files, highlighting unexpected security risks when chaining AI agent tools. The vulnerabilities were discovered and patched in late 2025, emphasizing the need for robust security in AI-integrated systems. <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>New AI coding model 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 speed. <a href="https://app.daily.dev/posts/eWdzokgG0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents are getting architectural upgrades</strong>: The A2A Java SDK 1.0.0.Alpha1 aligns with the forthcoming Agent2Agent Protocol 1.0, standardizing communication between AI agents, while new architectures emphasize modular, versioned ‘Agent Skills’ to prevent prompt bloat and improve scalability. <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>Google Gemini vulnerable to prompt injection via calendar invites</strong>: Security researchers found a vulnerability in Google Gemini where malicious calendar invitations could exploit the model’s automatic data ingestion to manipulate its behavior and exfiltrate private meeting data. <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>
<li><strong>Microsoft mandates Intune security updates for mobile apps</strong>: From January 19, 2026, Microsoft will enforce mandatory Intune Mobile Application Management security updates, requiring iOS and Android apps to update to the latest 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>
<li><strong>New EU decentralized vulnerability database launched</strong>: The GCVE has launched <a href="http://db.gcve.eu" target="_blank" rel="noopener nofollow">db.gcve.eu</a>, a publicly accessible, decentralized cybersecurity vulnerability database aimed at reducing European reliance on US-based systems like CVE, integrating data from over 25 sources with autonomous identifier assignment. <a href="https://app.daily.dev/posts/i2uCJ3RGE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>North Korean hackers target fintech developers via VS Code</strong>: North Korean threat actors are exploiting VS Code projects by distributing malicious job assessment repositories that, when opened, automatically deploy backdoors and malware on developers’ machines, targeting cryptocurrency and fintech sectors. <a href="https://app.daily.dev/posts/1wKLyk8gG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New tool streamlines cloud-native testing on Kubernetes</strong>: MetalBear released mirrord for CI, a new tool that allows CI tests to run directly on shared Kubernetes clusters, proxying traffic and environment variables to simulate production conditions without deploying code or setting up ephemeral environments. <a href="https://app.daily.dev/posts/s2plqXXTq" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI coding agents: 100x velocity or just 100x the hype?</strong>: Apparently, AI coding agents like Claude Code can deliver 10-100x velocity gains for ‘well-defined’ tasks, but struggle with anything ambiguous, often resorting to code repetition and omitting tests. Cursor even built a browser with hundreds of AI agents, which is ‘mostly correct’ but not production-ready. So, while the hype promises superhuman coding, the reality still requires ‘substantial domain expertise’ and ‘careful human task decomposition’ to avoid a mountain of technical debt, as noted by various developers and researchers. <a href="https://app.daily.dev/feed-by-ids?id=EqMZfdLoq&amp;id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5&amp;id=zWcZCkr3l" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vibecoding: the new executive superpower, or just more AI buzzwords?</strong>: Emergent, an Indian AI startup, just bagged $70 million for its ‘vibe-coding’ platform, claiming executives can now build full-stack apps without ‘advanced coding expertise.’ An opinion piece even suggests every exec should be ‘vibecoding’ to create hackathon platforms and internal tools. Because nothing says ‘cutting-edge’ like a term that sounds like it was invented in a corporate retreat brainstorming session, promising to turn every manager into a ‘no-code’ wizard, as if engineering teams just vanished. <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 is disrupting tech, but mostly by laying off UI component developers</strong>: Ryan Carniato predicts JavaScript frameworks will go ‘isomorphic-first’ and integrate async deeply, while Mark Thompson from Angular says frameworks don’t steal users, they optimize for their own. Then, Tailwind Labs lays off staff, blaming AI for reducing demand for pre-built UI components. So, AI isn’t just changing how we code; it’s apparently making entire job categories obsolete, proving that the ‘disruption’ is real, just maybe not where the hype machine initially pointed, as reported by various tech outlets. <a href="https://app.daily.dev/feed-by-ids?id=5NmTX91Eg&amp;id=bTgYZefU7" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
