<h2>Must know</h2>
<ul>
<li><strong>AI coding agents: the good, the bad, and the necessary guardrails</strong>: AI coding agents like Claude Code and GitHub Copilot promise significant velocity gains, with some reporting 10-100x speedups for well-defined tasks. However, they struggle with ambiguous requirements, can introduce more vulnerabilities, and often lead to technical debt if not carefully managed, according to developer accounts and research cited in multiple articles. The consensus, echoed by industry experts, is that strong human oversight, rigorous testing, automated quality checks, and clear architectural documentation are essential to harness AI’s benefits without sacrificing code quality or security. LLVM has even adopted a ‘human in the loop’ policy for AI-assisted contributions, requiring personal review and accountability. <a href="https://app.daily.dev/feed-by-ids?id=EqMZfdLoq&amp;id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5&amp;id=EdB2gt1Vu&amp;id=tr2l7aEc5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI accelerates exploit and malware development</strong>: The cybersecurity landscape is shifting dramatically as AI, specifically large language models like Opus 4.5 and GPT-5.2, can now generate 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. Check Point Research also identified VoidLink, the first known AI-generated malware suite, developed in under a week by a single person using AI agents—a task estimated to take 30 weeks for human teams. These developments, reported on January 20, 2026, indicate AI is industrializing offensive cyber capabilities, making human expertise in defense more critical than ever. <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 has cemented its role as the go-to infrastructure for enterprise AI, with 82% of organizations using it in production and 66% leveraging it for AI inference, according to the 2025 CNCF Annual Cloud Native Survey. However, the survey highlights a critical shift: cultural resistance, not technical complexity, is now the primary barrier to adoption. Mature organizations are distinguishing themselves through GitOps, platform engineering, and robust observability, indicating that successful AI integration is less about the tech and more about organizational readiness and practices. <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>New AI models for local coding and agentic tasks</strong>: GLM-4.7 Flash, a 30B-A3B mixture-of-experts model, is making waves for local AI coding tasks, outperforming competitors like Qwen 3-30B-A3B on benchmarks like SWBench and Tau Squared Bench, particularly in tool calling abilities. <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 European Union has launched <a href="http://db.gcve.eu" target="_blank" rel="noopener nofollow">db.gcve.eu</a>, a public, decentralized cybersecurity vulnerability database, aiming to reduce reliance on US-based systems and integrate 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 projects</strong>: North Korean threat actors are exploiting VS Code projects by distributing malicious job assessment repositories that, when opened, automatically deploy backdoors and malware, targeting developers in crypto, blockchain, and fintech sectors. <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>AI-driven ERP and modernization: hype vs. reality</strong>: While some studies claim AI-driven ERP can deliver 25% productivity gains, a PwC survey of 4,454 CEOs reveals 56% see no revenue increase or cost decrease from AI spending, highlighting a significant gap between hype and realized impact. <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>New Agent2Agent SDK and architectural patterns for AI agents</strong>: The A2A Java SDK 1.0.0.Alpha1 has been released, aligning with the forthcoming Agent2Agent Protocol 1.0, emphasizing standardized communication, modularity, and extensibility for AI agent architectures, moving towards skill-based systems to avoid prompt bloat. <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 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>FalkorDB claims massive speedup over Neo4j for graph queries</strong>: FalkorDB, a new graph database, claims to be 10-496x faster than Neo4j for multi-hop queries by implementing graphs as sparse matrices and leveraging linear algebra for parallel traversal, targeting real-time AI agents and RAG systems. <a href="https://app.daily.dev/feed-by-ids?id=GE8qOCt7y&amp;id=GHFDClpAg&amp;id=bbohKtF7X" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Vibecoding: the new executive superpower, apparently</strong>: Emergent, an Indian AI startup, just bagged $70 million for ‘vibe-coding,’ a concept that supposedly lets executives build full-stack apps without, you know, actual coding expertise. An opinion piece even suggests every exec should be ‘vibecoding’ to build hackathon platforms and metric tools. Because nothing says ‘robust enterprise solution’ like a system built on ‘vibes’ by someone who thinks ‘coding’ is just telling an AI what to do. What could possibly go wrong when the CEO decides to ‘vibe’ a new financial system? <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 memory leaks are a feature, not a bug</strong>: Turns out, AI-generated SwiftUI code is a master of creating persistent memory leaks by strongly capturing <code>self</code> in deeply nested closures. Engineers had to refactor by adding weak references and explicit capture lists—you know, basic best practices that apparently AI hasn’t quite grasped yet. So, while AI can churn out code at lightning speed, it seems we’re still stuck cleaning up its elementary mistakes. Progress! <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: where your calendar invites become prompt injection vectors</strong>: Security researchers found a delightful vulnerability in Google Gemini: malicious calendar invites can be used for prompt injection, allowing attackers to manipulate the model and exfiltrate private meeting data. Apparently, Gemini’s ‘automatic ingestion’ of event data means it’s happy to process embedded natural language instructions, even if they’re nefarious. So, next time you get a meeting invite, maybe double-check it’s not secretly telling your AI to spill your secrets. Zero Trust for your calendar, anyone? <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>
