<h2>Must know</h2>
<ul>
<li><strong>AI code generation: more hype than help?</strong>: Multiple engineering teams report that AI-generated code, while compiling, often lacks architectural context and leads to poor state management, ultimately slowing down projects due to extensive refactoring. ChatGPT, for instance, has been observed creating non-existent API elements, causing integration failures. This contradicts the hype, as a Google engineer’s claim that Claude Code could replicate a year’s work in an hour was later clarified as a ‘decent toy version,’ not production-ready, as reported by various industry analyses. <a href="https://app.daily.dev/feed-by-ids?id=STWE5tRXe&amp;id=UAphs9F5B&amp;id=euzGVQGzQ&amp;id=gMHZMigM0&amp;id=m6P4U0J0b&amp;id=wCQfiCXDa" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Major vendors ditch MITRE ATT&amp;CK evaluations</strong>: Microsoft, SentinelOne, and Palo Alto Networks have pulled out of the 2025 MITRE ATT&amp;CK Evaluations for endpoint detection and response, citing resource prioritization for product development. MITRE’s CTO admitted the 2025 test became more demanding, expanding to cloud environments and alert tracking. Industry experts are calling the process ‘vendor theater,’ suggesting it’s more for PR than genuine security improvement, according to CSO Online. <a href="https://app.daily.dev/feed-by-ids?id=c0ZYfkcgP&amp;id=hu11lcW1J" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Mandiant drops NTLMv1 rainbow tables, urges deprecation</strong>: Mandiant has publicly released rainbow tables capable of cracking Net-NTLMv1 password hashes in under 12 hours using consumer hardware. This move aims to force the deprecation of the 20-year-old, vulnerable protocol, which Mandiant says is still widely used despite Microsoft’s recommendations. They’ve provided tools and guidance for security teams to detect and disable NTLMv1, emphasizing the triviality of credential theft when organizations rely on it. <a href="https://app.daily.dev/posts/2q2phjvWS" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AI agents: old wine in new bottles?</strong>: Many sources suggest that ‘AI agents’ are largely a rebranding of existing API-centric automation, with real advancements focusing on building machine-readable APIs and semantic infrastructure rather than novel breakthroughs. <a href="https://app.daily.dev/feed-by-ids?id=4rJ96gwiU&amp;id=6IAZSWIz1&amp;id=6XeVWOBXO&amp;id=77jDjhI5C&amp;id=sH4hATTfN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA boosts AI gaming and streaming with DLSS 4.5</strong>: NVIDIA announced DLSS 4.5 with Dynamic Multi Frame Generation for up to 5 additional frames per rendered frame, enabling 240+ FPS path-traced gaming on RTX 50 Series GPUs, alongside new G-SYNC Pulsar monitors and expanded GeForce NOW access. <a href="https://app.daily.dev/posts/jXjzYY9vI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Specialized SLMs challenge large LLMs</strong>: Research suggests that specialized Small Language Models (SLMs) with as few as 7 million parameters can outperform large frontier LLMs on specific reasoning tasks, offering significant cost reductions and enabling a ‘digital factory’ model for AI. <a href="https://app.daily.dev/feed-by-ids?id=J1BKrq2u6&amp;id=J1IZlkzNv&amp;id=Yuaq0Gyje&amp;id=dliP781v1" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI data centers face skilled worker shortage</strong>: The boom in AI data center construction is creating a massive demand for skilled workers, with critical shortages in roles like electrical engineers and HVAC technicians, prompting companies and educational institutions to launch new training programs. <a href="https://app.daily.dev/posts/JE89HgTw1" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA DGX Spark and Station bring AI supercomputing to the desktop</strong>: NVIDIA unveiled DGX Spark and DGX Station, deskside AI supercomputers powered by Grace Blackwell architecture, allowing developers to run large-scale AI models locally with significant performance gains and memory optimizations. <a href="https://app.daily.dev/posts/bXoqeH8eP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare automates Salt debugging with Jetflow</strong>: Cloudflare developed Jetflow, an event-driven observability framework, to automate debugging of SaltStack configuration errors across millions of servers, reducing release delays and eliminating manual log analysis. <a href="https://app.daily.dev/posts/s63JTpz4h" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Chrome 145 beta brings CSS and Web API updates</strong>: Chrome 145 beta introduces new CSS enhancements like text-justify and column wrapping, along with Web API updates including Navigation API destination exposure and ECMAScript Map upsert methods. <a href="https://app.daily.dev/posts/oWgZOWIXh" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI coding assistants: the decade-old promise of developer replacement, still unfulfilled</strong>: Apparently, every decade since 1969, someone’s been trying to replace developers with tools like COBOL, CASE, low-code, and now AI coding assistants. Yet, here we are, still needing actual humans to untangle the mess. It seems the ‘complexity’ of software isn’t just about typing, but about the pesky intellectual work of reasoning through edge cases and system design, a concept apparently lost on those pushing the ‘AI will code everything’ narrative. <a href="https://app.daily.dev/posts/DSeK6EEPJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Claude Co-work: AGI in a sandboxed Ubuntu VM, with broken authentication</strong>: Anthropic’s new Claude Co-work desktop app is touted as a ‘small taste of AGI,’ promising to automate tasks for non-developers. However, early reports suggest it’s more of a ‘small taste of frustration,’ plagued by broken authentication, UI glitches, and confusing permission models. So, while it can organize files and control Chrome, you might need a human to actually log in and make it work. Progress! <a href="https://app.daily.dev/posts/OemgIV2rY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-hyped defenses: because infinite possibilities are just too much for AI</strong>: AppGuard is out here calling out AI-based malware detection as ‘fundamentally flawed’ because, get this, AI can’t parse ‘infinite possibilities.’ Instead, they’re pushing a ‘default-deny’ approach, which sounds suspiciously like what we used to call ‘good security practices’ before AI came along to complicate everything. Apparently, some things are just too complex for our silicon overlords, even if they can generate code that doesn’t quite work. <a href="https://app.daily.dev/posts/M7evoWVEa" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
