<h2>Must know</h2>
<ul>
<li><strong>Ladybird browser pivots to Rust with AI assistance</strong>: The Ladybird browser project is ditching Swift for Rust, primarily for memory safety, after losing a year on Swift due to poor C++ interop. 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, their JavaScript engine, achieved byte-for-byte identical output with zero regressions, though the C++ codebase remains the main development track for now. <a href="https://app.daily.dev/posts/bavH6r0ZC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Uber open-sources uForwarder Kafka consumer proxy</strong>: Uber has open-sourced uForwarder, a push-based Kafka consumer proxy designed to handle trillions of messages daily across over 1,000 services. This system centralizes offset management and simplifies consumer logic by replacing direct Kafka client implementations with a gRPC push interface. Key features include context-aware routing, an out-of-order commit tracker to prevent head-of-line blocking, and an adaptive auto-rebalancer, aiming to improve hardware utilization and reduce consumer lag. <a href="https://app.daily.dev/posts/NOSp6Xv2e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-augmented FortiGate firewall compromises reported by AWS</strong>: AWS detected a financially motivated, Russian-speaking threat actor compromising over 600 FortiGate firewall instances across 55+ countries. The attackers, described as unsophisticated, leveraged generative AI tools for reconnaissance, exploitation instructions, and scripting. They exploited exposed management interfaces, weak credentials, and lack of MFA, highlighting that basic security hygiene would have prevented most of these AI-assisted attacks. <a href="https://app.daily.dev/posts/azBzuZltn" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>FreeBSD’s Rust kernel support could be stable enough this year</strong>: FreeBSD’s Q4 2025 report suggests Rust kernel programming interfaces (KPIs) will be stable enough for developers to start experimenting with new kernel code by early 2026, aiming for compatibility with existing drivers and Rust conventions. <a href="https://app.daily.dev/posts/H5Wur3zjw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kubernetes 1.35 signals AI operating system role</strong>: Kubernetes v1.35 introduces features like workload-aware scheduling with gang scheduling and stable in-place Pod resource resize, strengthening its role as an operational substrate for AI/ML workloads. <a href="https://app.daily.dev/posts/mAyDHBOpw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub introduces Agent HQ for unified AI coding agents</strong>: GitHub announced Agent HQ, a unified platform for coding agents like Claude, Codex, and GitHub Copilot, allowing users to run and compare multiple frontier agents simultaneously within VS Code, GitHub, and soon Copilot CLI. <a href="https://app.daily.dev/posts/CxN7d4Xat" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cohere releases Aya Tiny multilingual models</strong>: Cohere launched the Aya Tiny model family, small (~3.3B parameter) multilingual language models supporting over 70 languages, designed for mobile devices and underserved linguistic communities. <a href="https://app.daily.dev/posts/QusoaclzE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Gusto’s universal document processing platform</strong>: Gusto replaced fragile, one-off document parsers with a Universal Document Processing (UDP) platform, using AI as a model-agnostic abstraction layer for ingestion, classification, extraction, validation, and mapping of millions of documents. <a href="https://app.daily.dev/posts/CDG9JvVKQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI responses API now supports WebSockets</strong>: The OpenAI Responses API now supports WebSockets, enabling AI agents to run 30-40% faster, particularly for frequent tool calls, with Cursor editor already upgrading users for up to 30% speed improvements. <a href="https://app.daily.dev/posts/CNQCNZ2N0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ZeroDayRAT mobile spyware emerges, bypassing MFA</strong>: ZeroDayRAT, a new mobile spyware sold on Telegram, targets Android 5-16 and iOS up to 26, offering capabilities like keylogging, SMS control (for MFA bypass), and credential theft, commoditizing nation-state-level surveillance. <a href="https://app.daily.dev/posts/Q0yMWpjwU" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic accuses Chinese labs of industrial-scale model distillation</strong>: Anthropic is publicly accusing Chinese AI labs of “industrial-scale distillation attacks” on its Claude models, claiming they created 24,000 fraudulent accounts to extract capabilities. This, coming from a company facing its own copyright lawsuits for training on scraped content, feels a bit like the pot calling the kettle black, doesn’t it? And conveniently, it’s framed as a national security threat, just as the US debates chip export controls. <a href="https://app.daily.dev/posts/YcZmCF704" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google DeepMind claims image generation breakthrough</strong>: Google DeepMind supposedly “just solved” a major problem in image generation with an “embarrassingly elegant” fix, according to some random Twitter user. No details, no official announcement, just pure, unadulterated hype. Sounds like another one of those breakthroughs that will remain a mystery until it’s quietly forgotten, or until we get another vague tweet about it. <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</strong>: OpenAI is now recommending SWE-bench Pro for AI coding evaluations because their previous benchmark, SWE-bench Verified, became “saturated” due to “test-design issues and contamination.” So, they’re moving the goalposts because their models got too good at cheating on the old test. Nothing to see here, just another AI benchmark that will inevitably be gamed until they need a new one. <a href="https://app.daily.dev/posts/oZGyDrjcX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
