<h2>Must know</h2>
<ul>
<li><strong>Anthropic leaks Claude Code source maps exposing internal agent architecture</strong>: Anthropic accidentally included source map files in a recent npm release, exposing over 500,000 lines of unobfuscated TypeScript. The leak, detailed by security researchers and mirrored on GitHub, reveals internal tools like ‘KAIROS’ for autonomous background operations and ‘undercover mode’ for hiding AI authorship in commits. It also uncovered anti-distillation mechanisms that inject fake tool definitions to poison competitor training data. <a href="https://app.daily.dev/feed-by-ids?id=mOipUZHX1&amp;id=7folff4vx&amp;id=cLSmwFJgK&amp;id=d39v889GW&amp;id=gmKbBNZUJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google releases ADK Go 1.0 for production AI agents</strong>: Google’s Agent Development Kit for Go has reached version 1.0, introducing a self-healing ‘Retry and Reflect’ plugin that automatically corrects failed tool calls. The framework integrates native OpenTelemetry for tracing model executions and a new Agent2Agent protocol for cross-language communication between Go, Java, and Python agents. It also includes built-in Human-in-the-Loop support to gate sensitive operations behind manual approval. <a href="https://app.daily.dev/feed-by-ids?id=7E16JJKCo&amp;id=uLGIp8OgK" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pretext library claims 500x speedup for web text layout</strong>: React pioneer Cheng Lou has released Pretext, a 15KB TypeScript library that performs text measurement and layout entirely outside the DOM. By bypassing the browser’s legacy 1990s-era sequential layout engine, it eliminates common performance bottlenecks like Slack’s scroll jumping and janky AI chat streaming. Demos show the engine handling hundreds of thousands of virtualized text boxes at 120fps without triggering browser reflows. <a href="https://app.daily.dev/posts/MEe3G5Jez" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>GoChat open sources a Discord scale real time backend in Go</strong>: A service-oriented architecture using NATS, ScyllaDB, and WebRTC for high-concurrency chat and voice. <a href="https://app.daily.dev/posts/I0zpEh1cy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Ollama integrates Apple MLX for faster local inference on Mac</strong>: The update leverages unified memory for the Qwen3.5-35B-A3B model, targeting local agentic workloads. <a href="https://app.daily.dev/posts/onA13BGmb" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>JetStream 3.0 benchmark suite overhauls WebAssembly performance testing</strong>: Developed by Apple, Google, and Mozilla, the suite now includes 12 new Wasm workloads covering Rust and Kotlin. <a href="https://app.daily.dev/posts/Ucbjrtf6l" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Red Hat Training Hub 0.4.0 adds Unsloth for 2x faster fine tuning</strong>: The library now supports LoRA and QLoRA with 70% less VRAM usage via a unified Python API. <a href="https://app.daily.dev/posts/NhDoHKuBF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Wayland proposes new fractional scaling protocol to fix rendering artifacts</strong>: The xx-fractional-scale-v2 proposal moves away from lossy integer coordinates to unscaled pixel coordinates. <a href="https://app.daily.dev/posts/CkCY0X6sU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Hyprmoncfg provides a zero dependency TUI for Hyprland in Go</strong>: A spatial canvas editor with hotplug daemon support and automatic configuration revert safety. <a href="https://app.daily.dev/posts/Ouw5MxHKg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Astro 6.0 debuts with experimental Rust compiler and live collections</strong>: The release features a refactored dev server and native Content Security Policy support. <a href="https://app.daily.dev/posts/dWn7iFESu" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The AI thinking token tax is a denial of wallet attack</strong>: A Stanford study found that ‘reasoning’ models like Gemini Flash and GPT-5.2 bill hidden thinking tokens at full output rates, making some ‘cheap’ models 28x more expensive in practice. Providers are essentially charging you for the AI’s internal monologue while adversaries can craft prompts to force expensive, infinite reasoning loops. It’s a brilliant business model: charge the customer for the model’s own indecision. <a href="https://app.daily.dev/posts/jF995M9kZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Microsoft Copilot adds a Council feature while stock tanks</strong>: As Microsoft faces its worst quarter since 2008 due to AI skepticism, they’ve launched ‘Council’ to let you compare different AI models side-by-side. It’s a classic move: when the underlying tech isn’t delivering the promised ROI, just add more UI so users can watch two different hallucinations fight in real-time. Maybe the ‘Critique’ feature can explain why the stock is down 25%. <a href="https://app.daily.dev/posts/TZFsy765M" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vibe coding is the new technical debt accelerator</strong>: The industry is rebranding ‘writing code you don’t understand’ as ‘vibe coding,’ with only 10% of AI-generated solutions actually being production-ready. We are effectively building a ‘subprime technical debt crisis’ where junior devs defer all complexity to future models that might not ever be smart enough to fix the mess. If you can’t explain how your agent solved the bug, you didn’t solve the bug; you just got lucky. <a href="https://app.daily.dev/feed-by-ids?id=osQspcxBm&amp;id=udI5FqrrI" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
