<h2>Must know</h2>
<ul>
<li><strong>Turbopack and Next.js 16 boost web platform performance</strong>: Next.js 16 has stabilized Turbopack as its default bundler, promising 2-5x faster production builds and 10x faster hot reloading. This update also introduces explicit ‘use cache’ directives for more predictable caching and AI-assisted debugging via Model Context Protocol (MCP) integration in DevTools, as reported by Syncfusion and Next.js. Turbopack’s speed comes from fine-grained incremental computation and file system caching, leveraging research from projects like Salsa and the Rust compiler. <a href="https://app.daily.dev/feed-by-ids?id=GHFDClpAg&amp;id=91xiEmcVe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding agents accelerate exploit development</strong>: LLMs like Opus 4.5 and GPT-5.2 are now generating sophisticated exploits for zero-day vulnerabilities, bypassing modern security mitigations such as ASLR and CFI. An experiment showed GPT-5.2 solving a complex exploit challenge in 3 hours for about $50, indicating that exploit development is becoming industrialized, limited by token throughput rather than human expertise, according to Hacker News. This shift means AI is directly enabling threat actors to scale offensive cyber operations. <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 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. These flaws, discovered in June 2025 and patched by December, highlight new security risks in AI-integrated systems, especially those involving code execution or file access. The exploits relied on manipulating AI agent workflows rather than direct system access. <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>AI-powered code reviewer built in 80 hours</strong>: A team built a GitHub-integrated, AI-powered code reviewer using AWS serverless and Amazon Bedrock in just 80 hours, highlighting that most effort went into handling API limits, retries, and webhooks rather than the AI component itself. <a href="https://app.daily.dev/posts/rTTGdX9v4" 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 like CVE. <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 project files to deploy backdoors and malware on developer machines, targeting those in cryptocurrency, blockchain, and fintech sectors. <a href="https://app.daily.dev/posts/1wKLyk8gG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>FalkorDB claims massive speedup over Neo4j for multi-hop queries</strong>: FalkorDB, a graph database using linear algebra and sparse matrices, claims to be 10-496x faster than Neo4j for multi-hop queries, targeting real-time AI agents and RAG systems. <a href="https://app.daily.dev/posts/GE8qOCt7y" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>WebAssembly advances with new features and broader adoption</strong>: WebAssembly saw significant progress in 2025 with Safari closing feature gaps, new proposals reaching standardization, and increased adoption across browsers and edge computing platforms. <a href="https://app.daily.dev/posts/jtf8mrjCJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated SwiftUI code causes memory leaks</strong>: AI-generated SwiftUI code led to persistent memory leaks by failing to use weak references in deeply nested closures, requiring manual refactoring to stabilize memory usage. <a href="https://app.daily.dev/posts/0Dxh8ahUn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft Intune mandates 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 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>Vibecoding: the new buzzword for ‘we still need humans to code’</strong>: Emergent, an Indian AI startup, just bagged $70 million for ‘vibe-coding,’ claiming 5 million users and $50 million ARR. Meanwhile, an opinion piece suggests executives should be ‘vibecoding’ to build apps without engineers. It’s almost like they’re trying to convince us that AI can replace skilled developers, but then they mention the need for ‘substantial domain expertise’ and ‘careful human task decomposition’ for AI coding agents to actually work. So, is it magic, or just another fancy term for ‘we still need competent people to guide the AI’? <a href="https://app.daily.dev/feed-by-ids?id=AllrxKqHW&amp;id=McILqauKu&amp;id=EqMZfdLoq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent autonomy: still mostly a pipe dream</strong>: Claims of AI agents independently delivering complex systems are still largely disputed. While Cursor’s AI agents built a 1 million-line web browser in Rust in a week, it’s ‘mostly correct’ but not production-ready and required direct referencing of standards. GitHub Copilot users also report 322% more code vulnerabilities. So, while AI can churn out code, the ‘autonomy’ part seems to require a lot of human babysitting, quality checks, and ‘intentional slowdowns to ensure quality.’ Sounds like a lot of extra steps for something that’s supposed to be autonomous. <a href="https://app.daily.dev/feed-by-ids?id=GPLZOgOgP&amp;id=J24eplQdr&amp;id=cSbwL1BS5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI hype cycle: rinse, repeat, disappoint</strong>: Apparently, we’re still stuck in the same old AI hype cycle: exaggerated claims, followed by measured expectations, then disappointment. Social media fuels unrealistic predictions, especially about software engineers becoming obsolete. Analysts warn that discussions are unbalanced, focusing on benefits while ignoring drawbacks. It’s almost as if the tech industry has a short-term memory problem when it comes to new technologies, constantly over-promising and under-delivering, only to repeat the cycle with the next shiny AI model. <a href="https://app.daily.dev/feed-by-ids?id=j0xF0WiVr&amp;id=yCkgCoa6x" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
