<h2>Must know</h2>
<ul>
<li>GitHub launches Agent HQ for unified AI coding: GitHub launched Agent HQ, a unified platform for coding agents like Claude, Codex, and GitHub Copilot, accessible with a single subscription. It allows users to run and compare multiple frontier agents simultaneously across various environments, including VS Code and GitHub Mobile. This platform aims to streamline AI-assisted development by enabling agents to autonomously open pull requests, respond to code reviews, and ship code, fostering an open ecosystem for custom agents. <a href="https://app.daily.dev/posts/CxN7d4Xat" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Ladybird browser pivots to Rust with AI assistance: Ladybird, an independent web browser project, is pivoting from C++ to Rust for memory safety, after a year-long detour with Swift proved unsuitable due to poor C++ interop and limited platform support. Lead developer Andreas Kling used AI agents like Claude Code and Codex to translate 25,000 lines of C++ to Rust in just two weeks, a task estimated to take months manually. The initial Rust port of LibJS, the browser’s JavaScript engine, achieved byte-for-byte identical output with zero test regressions, demonstrating AI’s potential for accelerating large-scale refactoring. <a href="https://app.daily.dev/feed-by-ids?id=bavH6r0ZC&amp;id=1QrlHS5ue" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Uber open-sources uForwarder Kafka consumer proxy: Uber open-sourced uForwarder, a push-based Kafka consumer proxy designed to handle trillions of messages daily across over 1,000 services. This system replaces direct Kafka consumer client implementations with a gRPC-driven push interface, centralizing offset management and simplifying consumer logic for microservices. Key features include context-aware routing, an out-of-order commit tracker to prevent head-of-line blocking, and an adaptive auto-rebalancer, all aimed at improving hardware utilization and reducing consumer lag. <a href="https://app.daily.dev/posts/NOSp6Xv2e" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li>OpenAI responses API now supports WebSockets: The OpenAI Responses API now supports WebSockets, which reportedly makes AI agents run 30-40% faster, especially for frequent tool calls, with Cursor editor already upgrading users to experience up to 30% speed improvements. <a href="https://app.daily.dev/posts/CNQCNZ2N0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>AI agent browser sandbox project announced: A new open-source project, ‘browser sandbox,’ provides a secure, zero-configuration browser environment for AI agents to safely interact with the web and execute bash, Python, and JavaScript commands. <a href="https://app.daily.dev/posts/AY0JUdVv6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Firefox 148 adds AI kill switch: Firefox 148 introduces an ‘AI kill switch’ under Settings &gt; AI Controls, allowing users to disable all AI-powered features or selectively block individual ones, with a promise that future updates will not re-enable disabled features. <a href="https://app.daily.dev/posts/UNAL37VrS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Cohere releases Aya Tiny multilingual models: Cohere released the Aya Tiny model family, small (~3.3B parameter) multilingual language models designed to support over 70 languages, including low-resource ones, and run on mobile devices. <a href="https://app.daily.dev/posts/QusoaclzE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>OpenAI releases gpt-realtime-1.5 model: OpenAI released gpt-realtime-1.5 through its Realtime API, strengthening voice workflows with more reliable instruction following, enhanced tool calling, and improved multilingual accuracy for low-latency, multimodal interactions. <a href="https://app.daily.dev/posts/6I1BavGOr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>jQuery v4 released with modernization focus: jQuery v4 was officially released, focusing on modernization by trimming legacy code, removing deprecated APIs, and dropping support for outdated browsers to create a leaner, faster, and more performant core library. <a href="https://app.daily.dev/posts/jtOuP3zQB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Reclaiming the web with personal hardware: A software engineer advocates for reclaiming the web from corporate control by owning personal hardware and using simple software like Markdown-based static websites and self-hosted servers on devices like Raspberry Pi. <a href="https://app.daily.dev/posts/lcNsx3UJj" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li>Anthropic cries foul over ‘industrial-scale distillation attacks’: Anthropic publicly accused Chinese AI labs DeepSeek, Moonshot AI, and MiniMax of “industrial-scale distillation attacks” on its Claude models, claiming they created over 24,000 fraudulent accounts to extract Claude’s capabilities. Anthropic frames this as a national security threat, conveniently coinciding with ongoing U.S. debates on AI chip export controls. Funny how these “national security threats” always seem to align with calls for stricter regulations that benefit the accusers, especially when Anthropic itself faces copyright lawsuits for training on scraped content. <a href="https://app.daily.dev/feed-by-ids?id=YcZmCF704&amp;id=s76VxSBCH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Google DeepMind ‘solves’ image generation problem: Google DeepMind reportedly “solved a significant problem in image generation” with a fix described as “embarrassingly elegant” by a Twitter user. No specific details, no official announcement, just a retweet with maximum hype. It’s almost as if they’re trying to generate buzz without actually, you know, explaining anything. Must be a new form of ‘elegant’ marketing. <a href="https://app.daily.dev/posts/fzJvtM3B4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li>Guide Labs debuts ‘inherently interpretable’ LLM: Guide Labs released Steerling-8B, an 8-billion-parameter LLM designed for “inherent interpretability,” tracing every token to human-understandable concepts and training data sources. They claim it achieves 90% of existing models’ capability using 2–7 times less training data. So, they’ve built a model that’s almost as good, but with a built-in explanation for its hallucinations. Because what we really need is an AI that can tell us why it’s confidently wrong. <a href="https://app.daily.dev/posts/0k3rLJK4y" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>