AI LABS

Jakarta EE

Kogan.com

Kotlin Weekly

Lobsters

Matt Wolfe

<h2>Must know</h2>
<ul>
<li><strong>AI agent security is a mess</strong>: AI coding agents, like Stripe’s ‘Minions’ generating 1,000+ PRs weekly, are prioritizing speed over security, creating a ‘security debt crisis.’ Research from Columbia University highlights issues like hardcoded API keys and public access policies. Meanwhile, a security researcher demonstrated ChatGPT recommending a malicious site, leading to a reverse C2 shell installation, attributing it to ‘ChatGPT SEO’ where LLMs trust self-declared legitimacy of malicious sites. <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 Google’s Gemini 3.1 Pro achieving top benchmark scores, including 78% on ARC AGI 2, it consistently underperformed Claude Opus 4.6 in real-world coding and UI generation tasks. Reviewers noted tool-calling failures, incomplete features, and poor environment detection, leading to ‘pure disappointment’ in practical applications. This suggests Google might be ‘benchmaxing’ at the expense of actual utility, a common pitfall in the AI race. <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 on a new workflow that allows Claude Code to convert production code into editable Figma designs. This enables developers to send a live web page to a Figma file with a simple prompt, facilitating design collaboration. The updated designs can then 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: a browser-native API for AI agents</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 orchestration and embedding infrastructure challenges</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 tasks, but challenges remain in agent observability and managing embedding model vendor lock-in. <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>LLM inference optimization for low-resource hardware</strong>: NTransformer, an open-source C++/CUDA engine, enables running Llama 3.1 70B on a single RTX 3090, while zclaw provides an open-source AI personal assistant firmware for ESP32 microcontrollers, showcasing efficient AI deployment across different hardware. <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>LangSmith Agent Builder’s memory system</strong>: LangSmith Agent Builder uses a filesystem-based memory system, storing agent knowledge as files in Postgres and exposing it as a virtual filesystem, allowing agents to evolve their understanding across sessions. <a href="https://app.daily.dev/posts/9xLrRFK2b" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong><a href="http://Lobste.rs" target="_blank" rel="noopener nofollow">Lobste.rs</a> struggles with SQLite migration</strong>: <a href="http://Lobste.rs" target="_blank" rel="noopener nofollow">Lobste.rs</a> attempted to migrate from MariaDB to SQLite but quickly reverted due to severe performance degradation, highlighting the challenges of database transitions for high-traffic sites. <a href="https://app.daily.dev/posts/RLQeRSeS0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Real-time rendering of millions of sparkles</strong>: Adobe Research and NVIDIA developed a new technique to render millions of microscopic glint particles in real-time at over 280 FPS on consumer GPUs, using a mathematical rule to generate particle positions on-the-fly. <a href="https://app.daily.dev/posts/rT496vKCA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Jakarta EE 12 Milestone 3 progresses</strong>: Jakarta EE 12 Milestone 3 is moving forward, with all specifications expected to update their parent pom.xml to the newly released EE4J Parent 2.0.0 for Maven Central staging, showcasing continued evolution for Enterprise Java. <a href="https://app.daily.dev/posts/xMxkpTx0S" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Orbital compute: the new AI battleground</strong>: Apparently, the AI race is no longer about algorithms but about who controls ‘orbital compute,’ with SpaceX-xAI leading the charge with their rockets, Starlink, and plans for a million-satellite orbital data center. Because, you know, terrestrial electricity grids and pesky permitting delays are just too much hassle for our AI overlords. Who needs ground when you have space, right? <a href="https://app.daily.dev/posts/QqzFDNqqS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Meta patents AI to post after you’re dead</strong>: Meta has patented an AI system that can simulate a deceased user’s social media activity, including posting and chatting, using their past data. Because nothing says ‘respect for the departed’ like having an AI chatbot impersonate them for eternity. Just what we needed: more digital ghosts haunting our feeds. <a href="https://app.daily.dev/posts/AeW0pHCcd" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>10 OpenClaw use cases everyone needs</strong>: An article touts OpenClaw, an autonomous AI agent, for everything from automated dependency maintenance to cloud cost monitoring and even end-to-end app building. The ‘key gotcha’? It can’t respond to permission prompts autonomously, so you either pre-configure everything or use a ‘dangerously-skip-permissions’ flag. Because what could possibly go wrong with an autonomous AI agent running with skipped permissions? <a href="https://app.daily.dev/posts/WCj2dFu7Q" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
