<h2>Must know</h2>
<ul>
<li><strong>AI coding agents are getting scary good, but still need you</strong>: New benchmarks show AI coding agents like Claude Opus 4.5 and GPT-5.2-Codex high are delivering impressive results on complex tasks, with Claude being fast and polished, and GPT-5.2 showing strong architectural reasoning. Cursor even used hundreds of AI agents to build a web browser in Rust in about a week, as reported by Cursor’s blog. However, these tools still struggle with ambiguous requirements and can introduce vulnerabilities, with GitHub Copilot users seeing 322% more code vulnerabilities. The consensus, backed by research and expert opinion, is that human oversight, strong testing, and clear architectural guidance remain critical for safe and effective AI-assisted development. <a href="https://app.daily.dev/feed-by-ids?id=EqMZfdLoq&amp;id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5&amp;id=1eWv43sPn&amp;id=20Yl6IWvm&amp;id=jHrHB1kVi&amp;id=yIClekHIh&amp;id=ymVBkKkOa" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI is industrializing exploit generation</strong>: Experiments show LLMs like Opus 4.5 and GPT-5.2 can generate dozens of 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 by a single person using AI agents in under a week—a task that would take human teams 30 weeks. This indicates a significant shift in offensive cybersecurity, where AI is now directly accelerating and scaling complex cyber operations, as detailed in reports from Hacker News and Check Point Research. <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>Kubernetes is now the standard for enterprise AI, but culture is the new bottleneck</strong>: Kubernetes production usage has hit 82%, with 66% of AI workloads running on it for inference, making it the de facto standard for enterprise AI infrastructure, according to the 2025 CNCF Annual Cloud Native Survey. However, the survey also reveals that cultural resistance, not technical complexity, is now the primary barrier to adoption. This means that while the tech is mature, organizations need to focus on fostering a culture that embraces cloud-native practices like GitOps and platform engineering to truly leverage Kubernetes for AI. <a href="https://app.daily.dev/posts/Lwsgn3lZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>LLVM and Oh My Zsh adopt ‘human in the loop’ for AI contributions</strong>: LLVM and Oh My Zsh are formalizing policies requiring human review and accountability for all AI-assisted code contributions, aiming to combat the rise of low-quality, AI-generated pull requests and ensure maintainer time is used effectively. <a href="https://app.daily.dev/feed-by-ids?id=EdB2gt1Vu&amp;id=tr2l7aEc5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Git MCP server had critical RCE vulnerabilities</strong>: Anthropic’s Git MCP server was found to have three critical vulnerabilities, including path validation bypass and argument injection, which could be chained via prompt injection to achieve remote code execution, highlighting new attack surfaces in AI-integrated systems. <a href="https://app.daily.dev/posts/U6A8HrCeq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>EU launches decentralized vulnerability database</strong>: The EU’s 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 reliance on US-based systems and provide autonomous vulnerability identifiers. <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 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 accessing corporate resources. <a href="https://app.daily.dev/posts/PNEoUht10" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Taiwan and US sign AI and semiconductor supply chain pact</strong>: Taiwan and the United States have finalized a strategic trade agreement focused on semiconductors and AI supply chains, with Taiwan committing $250 billion in investments and credit guarantees for US production, and Taiwanese chip makers like TSMC receiving lowered tariffs. <a href="https://app.daily.dev/posts/suaVSewS2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>North Korean hackers target fintech developers via VS Code projects</strong>: North Korean threat actors are exploiting VS Code project configuration files to target fintech developers, distributing malicious job assessment projects that deploy backdoors and malware upon opening, enabling remote code execution and cryptocurrency mining. <a href="https://app.daily.dev/feed-by-ids?id=1wKLyk8gG&amp;id=Q9M3WuQfH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New Agent2Agent SDK and modular AI agent architectures</strong>: A new Java SDK for the Agent2Agent Protocol 1.0 is out, standardizing communication between AI agents, while discussions highlight modular ‘Agent Skills’ as a solution to prompt bloat and scalability issues in complex AI systems. <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>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI investment payoff: CEOs still waiting for the magic</strong>: PwC surveyed 4,454 CEOs, and a whopping 56% reported no revenue increase or cost decrease from their AI spending. Only 12% saw both. Meanwhile, only 14% of workers use generative AI daily. So, while the industry keeps hyping ‘agentic AI’ and ‘AI readiness,’ most executives are still wondering where the ROI is. Maybe the magic happens after you spend another few billion? <a href="https://app.daily.dev/feed-by-ids?id=Kov24DV15&amp;id=ogb2Wd6xs&amp;id=rEdiBiFbV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated code: the gift that keeps on leaking</strong>: Turns out, AI-generated SwiftUI code is a master of memory leaks, creating retain cycles and unbounded memory growth by ‘capturing self strongly in deeply nested closures.’ Who knew AI could be so bad at basic memory management? Engineers had to manually refactor with weak references and explicit capture lists. So much for AI making our lives easier; it’s just creating new jobs for debugging its mess. <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>Vibecoding: because ‘coding’ wasn’t vague enough</strong>: Emergent, an Indian AI startup, just raised $70 million for ‘vibe-coding,’ claiming it lets users build full-stack apps without ‘advanced coding expertise.’ An opinion piece even suggests every exec should be ‘vibecoding’ to build hackathon platforms and metric tools. So, now we’re not just coding, we’re ‘vibecoding.’ Sounds like a great way to generate a lot of ‘vibes’ and very little production-ready software, all while making execs think they’re developers. <a href="https://app.daily.dev/feed-by-ids?id=AllrxKqHW&amp;id=McILqauKu" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
