<h2>Must know</h2>
<ul>
<li><strong>Ladybird browser pivots to Rust with AI assistance</strong>: Ladybird, the independent web browser project, is ditching Swift for Rust, primarily for memory safety. 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. This human-directed AI workflow, involving hundreds of prompts, achieved byte-for-byte identical output for critical components, demonstrating AI’s potential to accelerate 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><strong>Anthropic launches AI code security tool</strong>: Anthropic launched Claude Code Security, an AI-powered tool for vulnerability detection and patch suggestion, built on Claude Opus 4.6. It uses reasoning-based analysis to find business logic flaws, differing from traditional static analysis. While it found over 500 vulnerabilities in open-source codebases, security experts have mixed reactions, questioning the lack of false-positive rates and the risks of outsourcing security definitions to non-deterministic models. <a href="https://app.daily.dev/posts/bgcFYmVK1" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub launches agent HQ for unified AI coding</strong>: GitHub introduced Agent HQ, a unified platform for coding agents like Claude, Codex, and GitHub Copilot, accessible with a single Copilot Pro Plus or Enterprise subscription. This platform allows users to run and compare multiple frontier agents simultaneously, supporting custom agents and enabling them to autonomously open pull requests and respond to code reviews. It aims to streamline AI-assisted development workflows across various environments. <a href="https://app.daily.dev/posts/CxN7d4Xat" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Uber open-sources uForwarder kafka consumer proxy</strong>: Uber open-sourced uForwarder, a push-based Kafka consumer proxy designed to handle trillions of messages daily, centralizing offset management and simplifying consumer logic for microservices. <a href="https://app.daily.dev/posts/NOSp6Xv2e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI releases gpt-realtime-1.5 for voice workflows</strong>: OpenAI released gpt-realtime-1.5 via its Realtime API, focusing on strengthening voice workflows with improved instruction following, enhanced tool calling, and better multilingual accuracy for low-latency speech-to-speech interactions. <a href="https://app.daily.dev/posts/6I1BavGOr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI responses API gets websockets for faster AI agents</strong>: 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 for up to 30% speed gains. <a href="https://app.daily.dev/posts/CNQCNZ2N0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Guide labs debuts inherently interpretable LLM</strong>: 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, achieving competitive performance with less training data. <a href="https://app.daily.dev/posts/0k3rLJK4y" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cohere releases aya tiny multilingual models</strong>: Cohere released the Aya Tiny model family, small (~3.3B parameter) multilingual language models supporting over 70 languages, including low-resource ones, designed to run on mobile devices. <a href="https://app.daily.dev/posts/QusoaclzE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Firefox 148 adds AI kill switch</strong>: 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 Mozilla promising future updates won’t re-enable them. <a href="https://app.daily.dev/posts/UNAL37VrS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Krafton appoints chief AI officer for ‘AI first’ strategy</strong>: Krafton, publisher of PUBG, appointed Kangwook Lee as its first Chief AI Officer, signaling an ‘AI First’ strategy focused on enhancing gameplay, streamlining developer workflows, and long-term research including a new robotics entity. <a href="https://app.daily.dev/posts/IUxnxAIyr" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic cries foul over ‘industrial-scale’ model distillation</strong>: Anthropic is publicly accusing Chinese AI labs DeepSeek, Moonshot AI, and MiniMax of “industrial-scale distillation attacks” on its Claude models, claiming they created 24,000 fraudulent accounts to extract capabilities. Anthropic frames this as a national security threat, conveniently aligning with calls for stricter chip export controls. Because, you know, only <em>they</em> get to train on scraped data, not anyone else. <a href="https://app.daily.dev/feed-by-ids?id=YcZmCF704&amp;id=s76VxSBCH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google deepmind ‘solves’ image generation problem (allegedly)</strong>: Google DeepMind reportedly “solved a significant problem in image generation” with an “embarrassingly elegant” fix, according to some random Twitter user. No actual details, no official announcement, just pure, unadulterated hype. I’m sure it’s a groundbreaking, world-changing solution that will revolutionize everything, just like every other vague AI claim. <a href="https://app.daily.dev/posts/fzJvtM3B4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI updates AI coding evaluation standard (again)</strong>: OpenAI announced a change in its recommended standard for frontier coding evaluations, now pushing SWE-bench Pro because the previous benchmark, SWE-bench Verified, became “saturated due to test-design issues and contamination.” So, they moved the goalposts because their models got too good at the old game. Nothing to see here, just another AI benchmark that will inevitably be gamed. <a href="https://app.daily.dev/posts/oZGyDrjcX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
