<h2>Must know</h2>
<ul>
<li><strong>AI agents are everywhere, and so are their vulnerabilities</strong>: Anthropic’s new Claude Co-work, designed for desktop automation, carries an unpatched prompt injection vulnerability that allows attackers to exfiltrate sensitive files without user approval, as reported by PromptArmor researchers. This flaw, previously seen in Claude Code, highlights Anthropic’s continued reliance on user vigilance rather than technical mitigations. Meanwhile, ServiceNow rapidly patched a critical AI authentication flaw (CVE-2025-12420) that allowed unauthenticated attackers to impersonate any user and bypass MFA, demonstrating the severe risks of integrating AI agents with legacy systems, according to AppOmni. These incidents underscore the urgent need for robust security in AI-driven workflows, especially as AI agents gain broader access to enterprise data and systems. <a href="https://app.daily.dev/feed-by-ids?id=s11rN52dz&amp;id=wzAqd15ET&amp;id=jaKSkIh92&amp;id=rAHVvRPEj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes gets an AI-powered diagnostics assistant</strong>: AWS has released AIOps Sherlock, an open-source diagnostics assistant that uses a multi-agent AI architecture to troubleshoot Kubernetes workloads on Amazon EKS. This tool, built with Strands Agents SDK and Amazon Bedrock, aims to augment platform engineers and SREs by providing intelligent automation and expert-level analysis, as detailed in the AWS samples. It includes a complete evaluation framework with OpenTelemetry integration and Langfuse observability, showcasing how AI can be used to make Kubernetes clusters less ‘boring’ by proactively identifying and resolving issues. <a href="https://app.daily.dev/feed-by-ids?id=HuWoOJXv8&amp;id=QzUGYBGCW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare launches ‘Code Orange’ to prevent global outages</strong>: After experiencing multiple major outages in late 2025, Cloudflare has introduced ‘Code Orange: Fail Small,’ a new resilience plan. This initiative mandates phased configuration rollouts, improved failure-mode handling, and streamlined emergency response procedures to prevent cascading failures across its global network, as reported by InfoQ. The plan aims to enhance deployment processes with staged validations and automatic rollbacks, addressing risks from rapid global configuration propagation that previously caused widespread disruptions affecting major platforms like LinkedIn, Zoom, and Shopify. <a href="https://app.daily.dev/posts/OdNHxfcwg" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>ClickHouse raises $400M, acquires Langfuse for AI analytics</strong>: ClickHouse secured $400 million in Series D funding, valuing the company at $15 billion, and acquired Langfuse, an open-source LLM observability platform, to create a unified AI analytics platform for real-time observability and debugging of AI workloads. <a href="https://app.daily.dev/posts/9QbO7XqRY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Go-based LLM gateway Bifrost is 40x faster than LiteLLM</strong>: Bifrost, an open-source LLM gateway written in Go, offers 40x lower latency and 68% lower memory usage than LiteLLM, providing unified access to 15+ AI providers with features like semantic caching, automatic failover, and adaptive load balancing. <a href="https://app.daily.dev/posts/P8RYoHcNL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub Copilot gets repository-specific ‘memories’</strong>: GitHub Copilot now features an opt-in ‘memories’ system that automatically learns and documents coding preferences and project guidelines, aiming to improve code consistency, speed up onboarding, and automate enforcement of team standards. <a href="https://app.daily.dev/posts/nCmOnsQS3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google previews AI-based SQL generation in BigQuery</strong>: Google is rolling out ‘Comments to SQL’ in BigQuery, a new feature that uses AI to convert natural-language instructions within SQL comments into executable SQL queries, aiming to accelerate data analysis by minimizing manual coding. <a href="https://app.daily.dev/posts/5AcBUpc2i" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft open-sources XAML Studio for .NET UI prototyping</strong>: Microsoft has open-sourced XAML Studio under the .NET Foundation, providing a tool for rapid prototyping and fast feedback loops for XAML-based UI development, with community contributions now enabled via GitHub. <a href="https://app.daily.dev/posts/hUJrXK08w" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Node.js releases address multiple CVEs</strong>: Node.js released versions 20.20.0 LTS and 25.3.0, both addressing multiple security vulnerabilities, including fixes for TLSSocket error handling, permission model enforcement, and stack overflow exceptions. <a href="https://app.daily.dev/feed-by-ids?id=DwsoMgMKH&amp;id=WX4br4Zyw&amp;id=Yy5NFIIOM&amp;id=Zr0Dv9dwZ&amp;id=EyLXGRUe2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AWS European Sovereign Cloud is now available</strong>: AWS European Sovereign Cloud is now generally available, with infrastructure located entirely within the EU and operated by EU residents, offering enhanced data residency and operational autonomy for European customers. <a href="https://app.daily.dev/feed-by-ids?id=DTVSjPW7g&amp;id=B8P78hRXl" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI coding assistants: still more hype than help?</strong>: Remember all the talk about AI making us 10x developers? Well, a developer spent two days building an app with OpenAI’s Codex and GPT-5, and while it scaffolded a Spring Boot backend, the React Native code was ‘poor quality’ and required ‘extensive refactoring.’ It seems AI coding tools still need ‘deep technical knowledge to evaluate and fix generated code,’ often producing ‘suboptimal solutions that create maintenance debt.’ So much for replacing us, it’s just giving us more work, as reported by Awesome Coding. <a href="https://app.daily.dev/posts/DdiXGTqPN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Raspberry Pi AI HAT+ 2: more sizzle than steak?</strong>: Raspberry Pi’s new AI HAT+ 2, priced at $130, promises local LLM inference. But benchmarks show the Pi 5 CPU is actually faster for LLM inference than the dedicated Hailo 10H accelerator on the HAT+ 2. And with only 8GB RAM, it struggles with larger quantized models. It seems a 16GB Pi 5 might be a ‘more practical option’ for many tasks, making the HAT+ 2 more suitable for ‘development and prototyping’ than actual use, according to reviews. <a href="https://app.daily.dev/feed-by-ids?id=nbjHwDlw3&amp;id=jqT4w3IyJ&amp;id=kkiq3XN3f" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Small language models: the new ‘disruptor’ of LLMs?</strong>: Apparently, your ‘100 Billion Parameter Behemoth is a Liability.’ Enterprises are now supposedly favoring Small Language Models (SLMs) because LLMs are too expensive and slow. Benchmarks ‘show’ 7-million parameter models outperforming frontier LLMs on reasoning tasks while being ‘10,000 times smaller.’ This narrative, pushed by articles like those on InfoQ, conveniently overlooks the fact that SLMs are only good for ‘closed, well-defined workflows’ and still need LLMs for ‘complex reasoning.’ So, it’s not a replacement, just another tool in the box, but don’t let that stop the hype cycle! <a href="https://app.daily.dev/feed-by-ids?id=HX43N3MkE&amp;id=J1BKrq2u6" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
