<h2>Must know</h2>
<ul>
<li><strong>AI agent security is a mess</strong>: AI coding agents, while boosting productivity (Stripe’s “Minions” generate 1,000+ PRs weekly), are creating a “security debt crisis” by prioritizing speed over secure practices, according to Columbia University research. Issues range from hardcoded API keys to malicious backdoors in projects like FinTrust-AI, which exfiltrated environment variables. Mitigations include spec-driven prompting with OWASP Top 10 and robust CI/CD scanners. <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, poor usability</strong>: Despite achieving top benchmark scores, Google’s Gemini 3.1 Pro consistently underperformed Claude Opus 4.6 in real-world coding and UI generation tasks. Reviewers noted significant issues like tool-calling failures, incomplete features, and poor environment detection, leading to the conclusion that Google might be “benchmaxing” at the expense of practical application. This highlights a critical gap between theoretical performance and actual developer utility. <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 a new workflow where Claude Code can convert production code into editable Figma designs. This allows developers to send a live web page to a Figma file with a simple prompt, facilitating design collaboration and enabling updated designs to be round-tripped back to code. This integration aims to streamline the design-to-development process, as announced by Anthropic. <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: browser-native AI agent tools</strong>: Google’s WebMCP is a new browser-native API allowing websites to expose structured JavaScript tools for AI agents to invoke directly within the browser, improving determinism and reducing token overhead. <a href="https://app.daily.dev/posts/61GmU0Tcy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent memory systems are evolving</strong>: LangSmith Agent Builder uses a filesystem-based memory system, storing agent knowledge as files in Postgres, allowing agents to evolve their understanding across sessions with human-in-the-loop approval. <a href="https://app.daily.dev/posts/9xLrRFK2b" 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, can run Llama 3.1 70B on a single 24GB RTX 3090 by streaming model layers via PCIe and using a 3-tier adaptive caching system. <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>New tools for AI agent development</strong>: The AI agent ecosystem is growing with tools like <code>cocoindex-code</code> for semantic code search and Hugging Face Skills for reusable AI/ML task definitions, aiming to improve agent reliability and evaluation. <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>KittenTTS: ultra-compact text-to-speech models</strong>: KittenTTS offers ultra-compact text-to-speech models (down to 15M parameters) optimized for edge and browser deployment, enabling client-side TTS even with modest audio quality trade-offs. <a href="https://app.daily.dev/posts/aXA1rlMhE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Elixir language server reaches release candidate</strong>: Expert, the new official Elixir language server merging Lexical and Next LS, has released its first release candidate, bringing compiler diagnostics, completions, and refactoring to VSCode. <a href="https://app.daily.dev/posts/ItEtgwo7d" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>React tour app development with modern tools</strong>: A #100DaysOfCode project demonstrates building a tour application using React, Axios for data fetching, React Query for caching, Unsplash API for dynamic images, and Zustand for global state management. <a href="https://app.daily.dev/posts/sNDTxPsDJ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Orbital compute will reshape the global AI race</strong>: Apparently, the SpaceX-xAI merger means the AI race is now about who controls orbital infrastructure, not algorithms. Because, you know, terrestrial electricity grids are just too much of a bottleneck for our AI overlords. Never mind the actual engineering challenges of putting data centers in space; it’s all about “infrastructure is destiny,” according to some. <a href="https://app.daily.dev/posts/QqzFDNqqS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>10 OpenClaw use cases everyone needs</strong>: AI LABS is here to tell us that OpenClaw, an autonomous AI agent, can handle everything from automated dependency maintenance to building entire apps from a PRD. Just run it on a Mac Mini and watch the magic happen. Because who needs human developers when you have an AI that can coordinate with Claude Code and Nano Banana skills? Just remember to pre-configure permissions, because even our AI overlords can’t click a button. <a href="https://app.daily.dev/posts/WCj2dFu7Q" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Adobe and NVIDIA: 10,000,000 sparkles at 280 FPS</strong>: Adobe and NVIDIA have apparently cracked the code on rendering millions of microscopic sparkles at 280 FPS. Instead of storing every particle, they just generate them on-the-fly with a mathematical rule. Because what every developer really needs is more real-time glitter on their metallic paint and snow, right? It’s almost like they’re solving problems nobody asked for, but hey, it’s only 337 lines of code. <a href="https://app.daily.dev/posts/rT496vRCA" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
