<h2>Must know</h2>
<ul>
<li><strong>Orbital compute reshapes the global AI race</strong>: The SpaceX-xAI merger, valued at $1.25 trillion, signals a shift in the AI race from algorithmic competition to infrastructure control. SpaceX now commands a full vertical stack, including rocket manufacturing, a vast Starlink constellation, and the Colossus supercomputer, with plans for an orbital data center. This move is driven by the strategic advantages of orbital compute—continuous solar power, vacuum cooling, and freedom from national borders—to overcome terrestrial bottlenecks like electricity grids and permitting delays, as highlighted by an unknown source. <a href="https://app.daily.dev/posts/QqzFDNqqS" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent security and development trends</strong>: AI coding agents, while boosting productivity (Stripe’s Minions generate 1,000+ PRs weekly), introduce significant security risks. Research from Columbia University indicates these agents prioritize speed over security, leading to a ‘security debt crisis’ with issues like hardcoded API keys. Malicious backdoors in projects like FinTrust-AI demonstrate real-world threats, emphasizing the need for robust defense-in-depth architectures, including AI gateways and nonhuman identity management, as detailed across several sources. <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>Anthropic and Figma launch code-to-design workflow</strong>: Anthropic and Figma have partnered to create 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 and allowing updated designs to be round-tripped back to code. This integration streamlines 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>Database design starts with consequences, not tables</strong>: Effective database design prioritizes understanding business domain consequences and schema change rigor over immediately defining tables, emphasizing normalization, deliberate denormalization, and robust constraint usage. <a href="https://app.daily.dev/posts/fI73pTtiz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LLM inference optimization for single GPUs</strong>: NTransformer, an open-source C++/CUDA engine, enables running Llama 3.1 70B on a single RTX 3090 by streaming layers via PCIe and using adaptive caching, achieving significant speedups. <a href="https://app.daily.dev/posts/Ahx4zgQks" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LangSmith agent builder uses filesystem-based memory</strong>: LangSmith Agent Builder employs a filesystem-based memory system, storing agent knowledge as files in Postgres, to provide persistent, evolving knowledge for task-specific agents. <a href="https://app.daily.dev/posts/9xLrRFK2b" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PegaProx offers multi-cluster Proxmox VE management</strong>: PegaProx is an open-source, web-based dashboard for unified multi-cluster management of Proxmox VE, providing live metrics, VM lifecycle control, and advanced authentication features. <a href="https://app.daily.dev/posts/zaDpYuaxU" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Git’s magic files for repository control</strong>: A comprehensive guide details Git’s special committed files like .gitignore and .gitattributes, explaining their purpose, syntax, and impact on repository behavior across various platforms. <a href="https://app.daily.dev/posts/NIh4SPpHk" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Model adaptation: prompt-based techniques vs fine-tuning</strong>: Adapting LLMs involves either prompt-based techniques for inference-time context or fine-tuning for stable behavior, with a hybrid approach often yielding optimal results. <a href="https://app.daily.dev/posts/3uzAfcnhB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Jakarta EE 12 milestone 3 content outlined</strong>: Jakarta EE 12 Milestone 3 is progressing, with all specifications expected to update their parent pom.xml to the newly released EE4J Parent 2.0.0 for Maven Central staging. <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>Google’s Gemini 3.1 pro: high benchmarks, low usability</strong>: Google’s Gemini 3.1 Pro is apparently the ‘smartest model ever made’ according to some, boasting top-tier benchmark scores like 78% on ARC AGI 2. Yet, when put to actual coding and UI generation tasks against Claude Opus 4.6, it consistently face-planted, struggling with tool-calling failures and incomplete features. So, it’s brilliant on paper, just not, you know, in reality. Classic ‘benchmaxing’ at its finest. <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><a href="http://Lobste.rs" target="_blank" rel="noopener nofollow">Lobste.rs</a> migrates to SQLite, then immediately reverts</strong>: <a href="http://Lobste.rs" target="_blank" rel="noopener nofollow">Lobste.rs</a>, in a bold move, attempted to migrate its database from MariaDB to SQLite, briefly going read-only for the switch. The result? Severe performance degradation and an immediate rollback. Apparently, the team had considered PostgreSQL before this grand experiment. It’s almost as if database choices require a bit more foresight than a coin flip. <a href="https://app.daily.dev/posts/RLQeRSeS0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenClaw: the AI agent that can do everything (except respond to prompts)</strong>: OpenClaw, an autonomous AI agent tool, promises to handle a ‘wide range’ of developer tasks, from automated dependency maintenance to building entire apps from a PRD. It can even coordinate with other AIs! The only minor catch? It ‘cannot respond to Claude Code permission prompts autonomously.’ So, it’s autonomous, but only if you babysit it through the crucial security steps. Revolutionary. <a href="https://app.daily.dev/posts/WCj2dFu7Q" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
