<h2>Must know</h2>
<ul>
<li><strong>AI agent security is a mess</strong>: AI coding agents, while boosting productivity, are creating a ‘security debt crisis’ by prioritizing speed over secure practices, leading to vulnerabilities like hardcoded API keys and public access policies. Research from Columbia University highlights this trend, while a security researcher demonstrated ChatGPT recommending malicious sites, showcasing the risks of ‘ChatGPT SEO.’ Mitigations include spec-driven prompting with OWASP Top 10 and thorough code reviews. <a href="https://app.daily.dev/feed-by-ids?id=DD6LZnw0Q&amp;id=Et2d8uHu3&amp;id=LUO8dPKde&amp;id=RH1TYGIET&amp;id=UGo4Ea1t8&amp;id=bDJrJKMJX&amp;id=em7afQ5sm&amp;id=l0KLAjYGc&amp;id=ohhVPexbM&amp;id=vXBWwcwBk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google’s Gemini 3.1 pro: great benchmarks, terrible usability</strong>: Despite achieving top benchmark scores, Google’s Gemini 3.1 Pro consistently underperforms Claude Opus 4.6 in real-world coding and UI generation tasks. Issues include tool-calling failures, incomplete features, and poor environment detection, leading to significant practical usability problems for agentic workflows. This suggests Google might be ‘benchmaxing’ at the expense of actual developer experience, as noted in multiple analyses. <a href="https://app.daily.dev/feed-by-ids?id=bIAFUawfa&amp;id=mGa3eARS3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic and Figma team up for code-to-design workflow</strong>: Anthropic and Figma have collaborated to enable Claude Code to convert production code into editable Figma designs, allowing developers to send live web pages to Figma files with a simple prompt. This workflow facilitates design collaboration and allows updated designs to be round-tripped back to code, streamlining the iteration process between design and development. <a href="https://app.daily.dev/posts/aCaE5xJ9z" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>WebMCP enables browser-native AI agent tools</strong>: Google’s WebMCP is a browser-native API allowing websites to expose structured JavaScript tools for AI agents to invoke directly within the browser, reducing token overhead and improving determinism. <a href="https://app.daily.dev/posts/61GmU0Tcy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent orchestration and observability tools emerge</strong>: New tools like Mission Control (an open-source AI agent orchestration dashboard) and specialized observability primitives are emerging to manage and debug the non-deterministic nature of AI agents. <a href="https://app.daily.dev/feed-by-ids?id=LY7p8IODB&amp;id=ZmcjIu8ai&amp;id=frjHTf2Ap&amp;id=xhDI1zj4i" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NTransformer runs Llama 3.1 70B on a single RTX 3090</strong>: NTransformer, an open-source C++/CUDA LLM inference engine, enables running Llama 3.1 70B on a single RTX 3090 through adaptive caching and layer streaming, achieving significant speedups. <a href="https://app.daily.dev/feed-by-ids?id=Ahx4zgQks&amp;id=WoJMuyqAQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>KittenTTS offers ultra-compact TTS for edge devices</strong>: KittenTTS by Kitten ML provides a family of ultra-compact, CPU-optimized text-to-speech models (down to 15M parameters) designed for efficient deployment on edge devices and in browsers. <a href="https://app.daily.dev/posts/aXA1rlMhE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LangSmith agent builder uses filesystem-based memory</strong>: LangSmith Agent Builder implements a filesystem-based memory system for task-specific agents, storing persistent knowledge in Postgres and exposing it as a virtual filesystem to the agent. <a href="https://app.daily.dev/posts/9xLrRFK2b" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Elixir language server Expert reaches release candidate</strong>: Expert, the new official Elixir language server merging Lexical and Next LS, has released its first release candidate, focusing on stability, compiler diagnostics, and monorepo support. <a href="https://app.daily.dev/posts/ItEtgwo7d" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Go’s FFI now faster than GDScript and Rust for Godot</strong>: Go bindings for Godot via GDExtension have achieved FFI performance surpassing GDScript and ‘safe’ Rust, primarily due to call batching optimizations and Go 1.26 improvements. <a href="https://app.daily.dev/posts/j8AoMYJTv" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>OpenClaw promises to automate everything, with a catch</strong>: OpenClaw, an autonomous AI agent tool, claims it can handle everything from automated dependency maintenance to end-to-end app building. But wait, you still need to pre-configure permissions or use a ‘dangerously-skip-permissions’ flag because it can’t respond to prompts autonomously. So, it’s autonomous until it needs a human, which is, you know, the whole point of autonomy. <a href="https://app.daily.dev/posts/WCj2dFu7Q" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong><a href="http://Lobste.rs" target="_blank" rel="noopener nofollow">Lobste.rs</a> attempts SQLite migration, immediately fails</strong>: <a href="http://Lobste.rs" target="_blank" rel="noopener nofollow">Lobste.rs</a> tried to migrate from MariaDB to SQLite, briefly went read-only, and then immediately reverted due to ‘severe performance degradation.’ It’s almost like swapping out a robust database for a file-based one without extensive testing might not be the best idea for a high-traffic site. But hey, at least they tried, right? <a href="https://app.daily.dev/posts/RLQeRSe20" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>SwiftUI’s GeometryReader: the gift that keeps on breaking</strong>: Apparently, using <code>GeometryReader</code> in every SwiftUI <code>List</code> row or wrapping your root view with it to get screen size is a recipe for disaster, causing CPU spikes, dropped frames, and broken layouts when the keyboard appears. It’s a classic case of developers using a tool for something it’s clearly not designed for, then wondering why their app is a janky mess. Just stop using it as a screen size API, folks. <a href="https://app.daily.dev/feed-by-ids?id=dGEGPWHMv&amp;id=pxYDPz0nO&amp;id=p71tzkvLl" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
