AI LABS

ExplainingComputers

Kogan.com

Kotlin Weekly

LangChain

Lobsters

<h2>Must know</h2>
<ul>
<li><strong>Anthropic and Figma bridge code to design with Claude</strong>: Anthropic and Figma teamed up to create a workflow that converts production code into editable Figma designs using Claude Code. This allows developers to send a live web page to Figma with a simple prompt, facilitating design collaboration and enabling round-tripping updated designs back to code. This is a significant step for AI-driven software development workflows, bridging the gap between design and code. <a href="https://app.daily.dev/posts/aCaE5xJ9z" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Gemini 3.1 Pro: high benchmarks, low usability</strong>: Google’s Gemini 3.1 Pro, despite boasting impressive benchmark scores, is reportedly falling short in real-world coding and UI generation tasks compared to Claude Opus 4.6. Issues like tool-calling failures, incomplete features, and poor environment detection plague its practical usability, leading some to suggest Google is ‘benchmaxing’ rather than focusing on reliable agentic task completion. This highlights a critical challenge in AI model development: benchmarks don’t always translate to practical performance. <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>AI agents are creating a security debt crisis</strong>: AI agents are introducing a ‘security debt crisis’ by prioritizing speed over secure practices, leading to vulnerabilities like hardcoded API keys and public access policies. Stripe’s ‘Minions’ autonomously generate thousands of pull requests weekly, showcasing the scale of AI-driven development, but also the potential for widespread security issues if not properly managed. 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>
</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 AI agents to directly invoke structured JavaScript tools 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>NTransformer runs Llama 3.1 70B on a single GPU</strong>: NTransformer, an open-source C++/CUDA engine, allows running Llama 3.1 70B on a single RTX 3090 by streaming 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>Embedding model deprecation causes vendor lock-in</strong>: The deprecation of Google’s <code>text-embedding-004</code> model highlights the vendor lock-in risk of persisted embeddings, requiring organizations to design for change in their embedding infrastructure. <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>LangSmith Agent Builder uses filesystem-based memory</strong>: LangSmith Agent Builder uses a filesystem-based memory system, storing memory as files in Postgres but exposing it as a virtual filesystem to agents for persistent, evolving knowledge. <a href="https://app.daily.dev/posts/9xLrRFK2b" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>KittenTTS brings ultra-compact TTS to the edge</strong>: KittenTTS offers ultra-compact text-to-speech models (down to 15M parameters) optimized for edge and browser deployment, making lightweight TTS possible client-side. <a href="https://app.daily.dev/posts/aXA1rlMhE" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Adobe and NVIDIA achieve real-time sparkle rendering</strong>: A new research technique from Adobe and NVIDIA enables real-time rendering of millions of microscopic glint particles at over 280 FPS on consumer GPUs by generating particle positions on-the-fly. <a href="https://app.daily.dev/posts/rT496vKCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Database design starts with consequences, not tables</strong>: Good database design starts with understanding business consequences, emphasizing schema changes, deliberate denormalization, proper indexing, and using constraints to encode business rules. <a href="https://app.daily.dev/posts/fI73pTtiz" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Meta patents AI to post after you’re dead</strong>: Meta has patented an AI system to simulate deceased users’ social media activity, including posting and chatting, using their past data. Because what we all really need is our dead relatives’ AI ghosts spamming our feeds, ensuring we can never truly escape the digital afterlife. As if social media wasn’t creepy enough already. <a href="https://app.daily.dev/posts/AeW0pHCcd" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenClaw: the ‘autonomous’ AI agent that still needs you</strong>: OpenClaw, an ‘autonomous AI agent tool,’ promises to handle everything from automated dependency maintenance to building entire apps from a PRD. It’s like a digital butler, but for your code. Just remember, it can’t respond to Claude Code permission prompts autonomously, so you still have to babysit it. So much for ‘autonomous.’ <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>’ SQLite migration: a brief, failed experiment</strong>: <a href="http://Lobste.rs" target="_blank" rel="noopener nofollow">Lobste.rs</a> tried to migrate from MariaDB to SQLite and immediately rolled back due to ‘severe performance degradation.’ They’re planning to retry after addressing the issues. It’s always fun to watch someone try to swap out a battle-tested database for something lighter, only to discover that performance isn’t just a suggestion. Good luck with that ‘temporary’ rollback. <a href="https://app.daily.dev/posts/RLQeRSeS0" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
