<h2>Must know</h2>
<ul>
<li><strong>Google and Moonshot AI launch new agentic models</strong>: Google introduced Agentic Vision for Gemini 3 Flash, transforming static image processing into interactive investigation with visual reasoning and code execution, available through Google AI Studio and Vertex AI. Moonshot AI released Kimi K2.5, an open-source visual agentic model for coding and research automation, featuring a Mixture of Experts architecture and an Agent Swarm capability, as reported by Google and Moonshot AI themselves. These developments aim to make visual AI practical for production systems and enhance AI agent coordination. <a href="https://app.daily.dev/feed-by-ids?id=KIgS0LnaZ&amp;id=2GP3EQ58p" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents struggle in enterprise production</strong>: Despite significant investment, enterprise AI agent implementations are facing challenges transitioning from lab environments to production, with only the largest companies seeing success. Redis CEO notes fewer production AI agents than anticipated, primarily due to difficulties in aggregating unstructured data and replicating human judgment for nuanced decisions. This suggests a ‘trough of disillusionment’ for AI ROI, even as major players continue to invest in these platforms as next-gen infrastructure. <a href="https://app.daily.dev/posts/7PYYjzg77" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical WinRAR vulnerability actively exploited</strong>: A critical WinRAR vulnerability, CVE-2025-8088, remains actively exploited by various threat actors, including state-sponsored groups and cybercriminals, despite a patch released in July 2025. This path traversal flaw allows attackers to hide malware using Windows Alternate Data Streams, facilitating the deployment of Remote Access Trojans (RATs). The widespread exploitation highlights the persistent danger of unpatched or improperly patched software, even months after a fix is available. <a href="https://app.daily.dev/posts/J0xyRLNez" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Google BigQuery integrates SQL-native AI inference</strong>: Google BigQuery now supports SQL-native managed inference, allowing data teams to deploy and run over 180,000 Hugging Face and Vertex AI models directly via SQL statements, eliminating the need for separate ML infrastructure. <a href="https://app.daily.dev/posts/eM9NtN7mD" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft updates Quantum Development Kit</strong>: Microsoft announced major updates to its Quantum Development Kit (QDK), including integration with VS Code and GitHub Copilot for AI-assisted quantum programming, and new specialized toolkits for quantum chemistry and error correction. <a href="https://app.daily.dev/posts/0i847vEah" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Dynatrace launches AI observability app</strong>: Dynatrace released its AI Observability app, supporting agentic AI frameworks like Amazon Bedrock AgentCore and LangChain Agents, providing end-to-end visibility for agent interactions, tool usage, and cost tracking across major cloud providers. <a href="https://app.daily.dev/posts/d1KKTrRCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>SERA open-source coding agent announced</strong>: SERA, an open-source coding agent, achieves state-of-the-art performance on SWE-Bench Verified at 54.2% and is affordable to train, matching or exceeding 100B+ parameter models on specific repositories. <a href="https://app.daily.dev/posts/y5jjtMeVw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rust Foundation releases 2025 annual report and strategic plan</strong>: The Rust Foundation released its 2025 Annual Report and a 2026-2028 strategic plan focusing on stable infrastructure, sustainable maintainer support, adoption growth, organizational engagement, and building global communities. <a href="https://app.daily.dev/posts/DRpplNBlF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Firefox and Windows 11 release updates</strong>: Firefox 147 shipped with major web platform features like Navigation API and CSS Anchor Positioning, while Windows 11 Insider Preview Build 26300.7674 includes fixes for File Explorer and Start menu issues. <a href="https://app.daily.dev/feed-by-ids?id=MHu1Kt5o3&amp;id=P4NFiRPo5&amp;id=YpmFb9dA2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google releases TranslateGemma open-source models</strong>: Google launched TranslateGemma, an open-source translation model suite built on Gemma 3 architecture, with 4B, 12B, and 27B parameter variants supporting 55 languages and outperforming baselines on WMT24++ benchmarks. <a href="https://app.daily.dev/posts/Fvt1eqDgo" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic CEO argues for minimal AI regulation, citing theoretical superintelligence</strong>: Anthropic CEO Dario Amodei penned a 22,000-word essay arguing for minimal AI regulation, focusing on theoretical superintelligent AI risks and advocating for restricting China’s chip access. Critics, however, suggest this is just ‘typical corporate resistance’ to regulation, conveniently ignoring present concerns like wealth inequality and job displacement, especially since Anthropic isn’t expected to be profitable until 2028. Apparently, the best way to avoid accountability for current AI issues is to distract everyone with sci-fi scenarios. <a href="https://app.daily.dev/posts/amOI7bITg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes ‘intended behavior’ allows full cluster compromise</strong>: A security researcher found an exploit in Kubernetes’ <code>nodes/proxy GET</code> call that allows full cluster compromise and privileged command execution without audit trails. The Kubernetes team, in their infinite wisdom, classified this as ‘intended behavior,’ not a bug. So, apparently, a feature that lets anyone take over your cluster silently is just how things are supposed to work. Good to know our ‘secure by design’ systems are so thoughtfully constructed. <a href="https://app.daily.dev/posts/bOiz0Chx6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered polymorphic phishing bypasses traditional security</strong>: A new polymorphic phishing attack uses AI models like DeepSeek and Gemini to dynamically generate malicious JavaScript directly in victims’ browsers, bypassing traditional security by reassembling malware client-side from trusted LLM domains. This ‘last mile reassembly’ is touted as a significant technical development in cyber threats. Because what we really needed was AI making phishing even harder to detect. User awareness, they say, is still critical. Right, because users are totally going to spot AI-generated, fragmented malware. <a href="https://app.daily.dev/posts/jfmiKNwaJ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
