<h2>Must know</h2>
<ul>
<li><strong>AI coding agents are getting scary good, but still need adult supervision</strong>: New benchmarks show GLM-4.7 Flash outperforming competitors in coding and agentic AI tasks, especially for local AI coding. Meanwhile, Cursor used hundreds of AI agents to build a web browser in Rust, generating over a million lines of code in a week, as reported by Cursor’s blog. However, research cited in J24eplQdr indicates that while tools like GitHub Copilot boost individual velocity, they also lead to 322% more code vulnerabilities and lower delivery stability. The takeaway: AI can write a lot of code, fast, but human oversight is still critical to avoid a security and quality nightmare. <a href="https://app.daily.dev/feed-by-ids?id=eWdzokgG0&amp;id=EqMZfdLoq&amp;id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5" target="_blank" rel="noopener nofollow">Read more</a></li>
<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%, with 66% of AI workloads running on it for inference. It’s officially the standard for enterprise AI infrastructure. But here’s the kicker: the survey, widely cited, also states that cultural resistance, not technical complexity, is now the main barrier to adoption. So, your tech is ready, but your people might not be. <a href="https://app.daily.dev/posts/Lwsgn3lZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-driven exploit development is here, and it’s industrializing fast</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 like ASLR and CFI. GPT-5.2 even solved a complex scenario in 3 hours for about $50, according to Hacker News. 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. This means the bottleneck for exploit generation is shifting from human expertise to LLM token throughput, accelerating the cyber arms race. <a href="https://app.daily.dev/feed-by-ids?id=WkT9zuzY1&amp;id=nLTC7GTyu" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>New AI agent frameworks and tools emerge for structured development</strong>: The A2A Java SDK 1.0.0.Alpha1 is out, aligning with the Agent2Agent Protocol 1.0 for standardized AI agent communication, while new tools like Vercel’s ‘skills’ CLI and Neuron for PHP provide structured ways to build and manage AI agent capabilities. <a href="https://app.daily.dev/feed-by-ids?id=LmGT7Q5vx&amp;id=PMa93QJZs&amp;id=ZcWzztsx0&amp;id=JdhS5Ojd6&amp;id=eqhBE21dP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Open-source AI coding agent OpenCode offers flexibility</strong>: OpenCode is an open-source AI agent for terminal-based coding tasks, offering more flexibility and avoiding vendor lock-in compared to proprietary alternatives like Anthropic’s Claude. <a href="https://app.daily.dev/posts/NR4LrQXl3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered observability tools for OpenShift AI environments</strong>: Red Hat released an AI-powered observability tool that translates complex infrastructure metrics into conversational insights for OpenShift AI, allowing natural language queries about model performance and cluster health. <a href="https://app.daily.dev/posts/jpKDLQQHM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Temporal and Braintrust integrate for observable AI agents</strong>: Temporal and Braintrust have integrated to provide durable workflow execution and LLM observability for AI agents, enabling production-ready agents with tracing, prompt versioning, and cost metrics. <a href="https://app.daily.dev/posts/vZ2X1Grfn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub enhances code-to-cloud traceability for supply chain security</strong>: GitHub now offers artifact metadata APIs for code-to-cloud traceability, linking build artifacts to source repositories and deployment contexts, supporting SLSA Build Level 3 security. <a href="https://app.daily.dev/posts/9XGazeUyY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New terminal SQL client with AI-powered query generation</strong>: SQDesk CLI is a new terminal-based SQL client supporting PostgreSQL, MySQL, and SQLite, featuring AI-powered query generation and refactoring, syntax highlighting, and auto-completion. <a href="https://app.daily.dev/posts/da02fw80D" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft Intune mandates security updates for mobile apps</strong>: Microsoft is enforcing 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 being blocked from 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>AI-powered code review: 80 hours to build, 90% non-AI work</strong>: A team built an AI-powered code reviewer in 80 hours, only to find that the AI component was about 10% of the work. The real challenge? API rate limits, latency, hallucinations, and the fact that most LLMs can’t reason beyond 32KB despite claiming millions of tokens. So much for AI doing all the heavy lifting, seems like we’re still stuck with the boring bits. <a href="https://app.daily.dev/posts/rTTGdX9v4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vibecoding: the new executive superpower, or just more AI hype?</strong>: Emergent, an Indian AI startup, just raised $70 million for ‘vibecoding,’ claiming executives can now build full-stack apps without advanced coding. An opinion piece even suggests every exec should be ‘vibecoding’ to build hackathon platforms and metric tools. Because nothing says ‘thoughtful architectural design’ like a CEO ‘vibecoding’ an internal tool, probably with zero tests and a mountain of technical debt. <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>LLVM and Oh My Zsh: AI contributions need human accountability, who knew?</strong>: LLVM and Oh My Zsh maintainers are now formally requiring human oversight for AI-assisted code contributions, citing an increase in low-quality, hard-to-reason-about PRs. Apparently, AI-generated code still needs a human to actually understand it and take responsibility. Shocking, I know, that ‘human in the loop’ isn’t just a buzzword for liability. <a href="https://app.daily.dev/feed-by-ids?id=EdB2gt1Vu&amp;id=tr2l7aEc5" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
