<h2>Must know</h2>
<ul>
<li><strong>Critical security flaw in ServiceNow AI agent integration</strong>: AppOmni discovered ‘BodySnatcher’ (CVE-2025-12420), a critical ServiceNow vulnerability allowing unauthenticated attackers to impersonate users with just an email, bypassing MFA and SSO. This could lead to rogue AI agents with escalated privileges accessing external environments, as reported by AppOmni. ServiceNow has confirmed the flaw and released an immediate patch, but the incident highlights the growing attack surface with enterprise AI adoption. <a href="https://app.daily.dev/posts/wzAqd15ET" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>US agencies push post-quantum security transition</strong>: U.S. federal agencies, including the FBI, NIST, and CISA, are moving beyond awareness to active implementation of post-quantum security measures. At the YQS2026 conference, new cryptographic migration strategies were detailed to counter ‘harvest now, decrypt later’ attacks, according to official government guidance. NIST has released initial standards, recommending hybrid cryptographic methods as interim solutions for this multi-year transition. <a href="https://app.daily.dev/posts/lRZsyJgWP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s Claude Code gets new features and API restrictions</strong>: Anthropic released Cowork, a research-preview feature for Claude Code, enabling automation of enterprise tasks beyond software development, such as summarizing calendars and creating reports. However, Anthropic has also blocked the use of API credentials from all subscription plans for third-party applications, limiting Claude access to their own tools, a move that has caused significant backlash in the developer community, as noted by developer reports and GitHub complaints. <a href="https://app.daily.dev/feed-by-ids?id=GMZHoSF60&amp;id=QvZHdydfi&amp;id=U2HzInRUP&amp;id=ejPwSyHLL" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>React frameworks converge, AI integration on the horizon</strong>: At React Conf, experts agreed that mainstream React frameworks are converging in features, making developer choice largely risk-free, with all major frameworks exploring AI integration via Model Context Protocol (MCP) endpoints. <a href="https://app.daily.dev/feed-by-ids?id=Ii8I7s3X0&amp;id=oWHYd6BqQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Livewire 4 introduces single-file components and performance boosts</strong>: Livewire 4 is out, bringing single-file components, an ‘islands’ feature for isolated rendering to improve performance, built-in drag-and-drop, and smooth transitions via the View Transitions API, all while maintaining backward compatibility. <a href="https://app.daily.dev/posts/OxhgZzaoQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>CrowdStrike acquires Seraphic Security for browser threat protection</strong>: CrowdStrike is buying Seraphic Security for $400 million to integrate its enterprise browser security technology, which isolates JavaScript engines and randomizes memory addresses, into the Falcon platform for enhanced real-time threat detection. <a href="https://app.daily.dev/posts/AKJAUyPZD" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>CNCF graduates Dragonfly, MemOS launches for LLM memory management</strong>: The CNCF has graduated Dragonfly, a peer-to-peer image and file distribution platform for container images and AI models, while MemOS launched as an open-source memory operating system for LLMs, claiming significant accuracy improvements and token savings. <a href="https://app.daily.dev/feed-by-ids?id=0rVfXzPjQ&amp;id=bjqQdB8fJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Firefox 147 enhances privacy and platform support</strong>: Firefox 147 is here with automatic Picture-in-Picture, reorganized tab settings, enhanced WebGPU support for Apple Silicon Macs, zero-copy hardware video playback for Windows AMD users, and stricter Enhanced Tracking Protection. <a href="https://app.daily.dev/posts/f7pzHnDhN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>BeaglePlay board gets fully open-source PowerVR graphics</strong>: The $99 BeaglePlay board now supports fully open-source, upstream PowerVR graphics drivers, a significant shift that addresses long-standing demands from the Linux and open-source communities for transparency and compatibility. <a href="https://app.daily.dev/feed-by-ids?id=HwIiwOf9Q&amp;id=NjVG9GX8g" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Agoda centralizes financial data with Apache Spark</strong>: Agoda consolidated its financial data pipelines into a single Apache Spark-based platform, FINUDP, to eliminate metric inconsistencies, processing millions of transactions daily with machine learning for anomaly detection, though it resulted in slower development velocity. <a href="https://app.daily.dev/feed-by-ids?id=SIYRILab7&amp;id=WuRILab7" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI agents porting code: the new 10x developer fantasy</strong>: Apparently, an experienced developer managed to port a Rust template engine to Go in 45 minutes of ‘active guidance’ and 10 hours of ‘autonomous agent work’ using AI. The article suggests this means AI agents are making cross-language ports feasible with ‘less effort.’ So, now we’re just going to let the bots rewrite everything, and our job is to babysit them? Sounds like a recipe for a whole new class of ‘idiomatic’ bugs. <a href="https://app.daily.dev/posts/VOMBPE1Xs" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI solves math problems, but still needs human babysitters</strong>: OpenAI’s GPT 5.2 is credited with solving 11 of 15 newly resolved Erdős problems, and ChatGPT is now a ‘reasoning assistant’ in math research. But don’t worry, humans still need to ‘verify proofs’ and ‘manage complex cases’ because AI models aren’t ‘fully autonomous.’ So, AI can do the grunt work, but we still have to clean up its messes. Progress, I guess. <a href="https://app.daily.dev/feed-by-ids?id=8o6H9wDXV&amp;id=mhyD4ij4A" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated code: functionally correct, architecturally a dumpster fire</strong>: A senior Android developer reviewed AI-generated Kotlin code and found it ‘functionally correct’ but ‘architecturally problematic,’ with issues like unnecessary complexity, race conditions where none exist, and poor recomposition patterns. So, AI can write code that ‘works,’ but it’s apparently incapable of writing code that’s actually good. Guess we’ll all be spending more time refactoring AI’s brilliant solutions. <a href="https://app.daily.dev/posts/O3swuQWEt" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
