<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 test regressions, proving AI can significantly accelerate large-scale refactoring when human-directed. <a href="https://app.daily.dev/posts/bavH6r0ZC" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic accuses chinese AI labs of industrial-scale model distillation</strong>: Anthropic is publicly accusing Chinese AI labs DeepSeek, Moonshot AI, and MiniMax of creating over 24,000 fraudulent accounts to distill its Claude models, generating 16 million exchanges to extract capabilities like agentic reasoning and coding. Anthropic frames this as a national security threat, arguing distilled models lack safety guardrails and could be misused. This comes amidst US debates on AI chip export controls, with Anthropic using these claims to push for stricter restrictions, despite facing its own copyright infringement lawsuits for training on scraped content. <a href="https://app.daily.dev/posts/YcZmCF704" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Shopify graphql api adopts discriminated unions</strong>: Shopify’s GraphQL Admin API (2026-04) is introducing a new <code>SubscriptionBillingAttemptState</code> field that replaces multiple nullable fields with a discriminated union pattern. This change deprecates older, loosely-typed fields in favor of strongly-typed, state-scoped equivalents, significantly improving API ergonomics and self-documentation. This is a solid move for type safety and clarity in API design. <a href="https://app.daily.dev/posts/WcjzF6cXJ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Firefox 148 adds AI controls and web platform updates</strong>: Firefox 148 introduces an ‘AI kill switch’ allowing users to disable or selectively block AI-powered features like ChatGPT sidebar integrations and AI link previews, alongside web platform updates including Trusted Types API and Sanitizer API for XSS mitigation. <a href="https://app.daily.dev/posts/UNAL37VrS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub introduces agent HQ for unified AI coding agents</strong>: GitHub’s new Agent HQ provides a unified platform for coding agents, integrating Claude, Codex, and GitHub Copilot, allowing users to run and compare multiple frontier agents simultaneously with a single Copilot Pro Plus or Enterprise subscription. <a href="https://app.daily.dev/posts/CxN7d4Xat" 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, centralizing offset management and simplifying consumer logic for microservices with features like context-aware routing and adaptive auto-rebalancing. <a href="https://app.daily.dev/posts/NOSp6Xv2e" 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 and serve languages often overlooked by larger models. <a href="https://app.daily.dev/posts/QusoaclzE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TSSLint 3.0 final release with reduced dependencies</strong>: TSSLint 3.0, the final major release of the lightweight TypeScript semantic linting tool, removes the esbuild dependency, adds a TSL compatibility layer, and re-introduces <code>@tsslint/compat-eslint</code>, while operating as a TypeScript Language Server plugin for near-instant diagnostics. <a href="https://app.daily.dev/posts/314okFZnQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent browser sandbox project announced</strong>: A new open-source project, ‘browser sandbox,’ provides a secure, zero-configuration browser environment for AI agents to interact with the web, pre-loaded with ‘agent-browser’ and ‘Playwright’ for safe execution of bash, Python, and JavaScript commands. <a href="https://app.daily.dev/posts/AY0JUdVv6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>JQuery v4 released with modernization focus</strong>: jQuery v4 has been 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>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>OpenAI updates AI coding evaluation standard</strong>: OpenAI, in a move that screams ‘our old benchmark was too easy for us,’ now recommends SWE-bench Pro for AI coding evaluations, ditching SWE-bench Verified because it became ‘saturated’ and ‘contaminated.’ So, they moved the goalposts after their models got too good at the old game. Convenient, isn’t it? <a href="https://app.daily.dev/posts/oZGyDrjcX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google deepmind claims image generation breakthrough</strong>: Google DeepMind supposedly ‘just solved’ a ‘dirty problem’ in image generation with an ‘embarrassingly elegant’ fix, according to a highly enthusiastic tweet. Details? None. Official announcement? Crickets. Sounds like another classic case of social media hype without any actual substance, probably just a new way to make AI-generated hands look slightly less terrifying. <a href="https://app.daily.dev/posts/fzJvtM3B4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-augmented fortigate firewall compromises reported by AWS</strong>: AWS reports that a ‘technically unsophisticated’ Russian-speaking threat actor compromised over 600 FortiGate firewalls using generative AI tools for reconnaissance and exploit instructions. Apparently, AI can make even amateurs dangerous, but the real takeaway is that basic security hygiene like MFA and not exposing management interfaces would have stopped most of it. So, AI isn’t magic, it just helps bad actors be lazy, just like it helps some developers. <a href="https://app.daily.dev/posts/azBzuZltn" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
