<h2>Must know</h2>
<ul>
<li><strong>AI coding agents are fast but need human guardrails</strong>: AI coding agents like Claude Code and GitHub Copilot can deliver significant velocity gains (10-100x) for well-defined tasks, with Cursor’s AI agents building a 1M+ line web browser in Rust in about a week. However, they struggle with ambiguous requirements, often produce vulnerabilities, and require extensive human oversight, peer reviews, and strong test coverage to avoid technical debt, according to developer accounts and research cited in Hacker News. The paradox is that AI can only safely improve productivity with intentional slowdowns to ensure quality. <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>Kubernetes is now the standard for enterprise AI, but culture is the new bottleneck</strong>: Kubernetes production usage has hit 82% among surveyed organizations, with 66% of AI adopters using it for inference, establishing it as the standard for enterprise AI infrastructure. The 2025 CNCF Annual Cloud Native Survey, widely cited, reveals that cultural resistance, not technical complexity, is now the primary barrier to adoption. Mature organizations are distinguishing themselves with GitOps and platform engineering practices, while sustainability concerns are growing for open-source infrastructure due to increasing AI workloads. <a href="https://app.daily.dev/posts/Lwsgn3lZ5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-driven exploit development is industrializing</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 solved a complex scenario in 3 hours for about $50, indicating exploit generation is now limited by LLM token throughput, not human expertise, as reported on 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, a task estimated to take 30 weeks for multiple human teams. <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>Next.js 16 boosts performance with Turbopack and AI debugging</strong>: Next.js 16 stabilizes Turbopack as the default bundler, offering 2-5x faster production builds and 10x faster hot reloading, while also introducing AI-assisted debugging via Model Context Protocol (MCP) integration in DevTools for AI agents to suggest fixes. <a href="https://app.daily.dev/posts/91xiEmcVe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New Java SDK standardizes AI agent communication</strong>: The A2A Java SDK 1.0.0.Alpha1 has been released, aligning with the upcoming Agent2Agent Protocol 1.0 specification to standardize communication between AI agents, featuring migration to Java records, Protocol Buffers, and enhanced API documentation. <a href="https://app.daily.dev/posts/LmGT7Q5vx" 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 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>AI-powered SQL client for the terminal</strong>: SQDesk CLI is a new terminal-based SQL client supporting PostgreSQL, MySQL, and SQLite, featuring an 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>PHP 8.4 changes break <code>http_build_query()</code> for enums</strong>: A PHP 8.4 update to <code>http_build_query()</code> now serializes backed enums by value and throws errors for non-backed enums, causing breaking changes and prompting the PHP URI toolkit to offer new compatibility modes. <a href="https://app.daily.dev/posts/FEB00NeSK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New micro-framework for minimalist PHP development</strong>: Lighthouse is a new minimal PHP micro-framework designed for rapid development of MVPs and internal tools, featuring zero-configuration setup, built-in security, and integration with modern tools like HTMX and Pico.css. <a href="https://app.daily.dev/posts/bI3vfttQJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Godot developer showcases sword-fighting combat system</strong>: Developer dpolk_ created a sword-fighting combat system demo in Godot, inspired by SUPERHOT’s time mechanics, showcasing advanced combat mechanics without a concrete game plan yet. <a href="https://app.daily.dev/posts/GLI89wvU6" 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-powered ‘vibe-coding’ startup, just bagged $70 million in Series B funding, claiming $50 million ARR and 5 million users. Meanwhile, some ‘experts’ are pushing ‘vibecoding’ as the next big thing for executives, suggesting it lets senior leaders build full-stack apps without bothering those pesky engineers. Because, clearly, the biggest problem in tech is that executives can’t code their own hackathon platforms, not, say, actual product-market fit or sustainable engineering practices. <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 readiness is driving modernization, or so they say</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. Yet, we’re told ‘AI readiness’ is now the big driver for application modernization, with ‘agentic AI’ being the new hype cycle. It seems companies are still chasing the AI dragon, even when their own leaders admit it’s not paying off, all while ‘modernization factories’ churn out continuous updates to enable this elusive AI functionality. <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>LLMs are great, but only if you ignore their actual reasoning capabilities</strong>: A team built an AI-powered code reviewer in 80 hours, only to find that published context window sizes for LLMs are basically fiction; most models fail beyond 32KB despite claiming millions of tokens. The AI component was a mere 10% of the work, with the rest spent on handling API limits, retries, and hallucinations. So, while the hype machine touts AI’s reasoning prowess, the reality is we’re still building elaborate scaffolding just to make these models pretend to be smart. <a href="https://app.daily.dev/posts/rTTGdX9v4" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
