<h2>Must know</h2>
<ul>
<li><strong>OpenAI’s harness engineering automates code generation</strong>: OpenAI has introduced “Harness Engineering,” an internal AI-driven methodology that uses Codex agents to automate tasks like code generation and testing. A five-month internal experiment reportedly resulted in a beta product with a million lines of code generated without manual writing, shifting engineers’ roles to designing environments and providing feedback. Martin Fowler noted this as a valuable framing of AI-enabled software development. <a href="https://app.daily.dev/posts/dPCNsAxds" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic and Figma streamline design-to-code with claude</strong>: Anthropic and Figma have teamed up to integrate Claude Code, allowing developers to convert live production code into editable Figma designs. This workflow aims to streamline collaboration, enabling design changes within Figma to round-trip back to the codebase. The integration uses the Figma MCP server for data exchange, facilitating AI-powered code-to-design and design-to-code synchronization. <a href="https://app.daily.dev/posts/aCaE5xJ9z" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare R2 local uploads reduce latency</strong>: Cloudflare has launched “Local Uploads” for its R2 object storage, now in open beta. This feature aims to cut cross-region write latency by up to 75% by writing data to the nearest Cloudflare PoP and then asynchronously replicating it to the bucket’s home region. Objects are immediately accessible, and there are no additional costs beyond standard Class A operations. <a href="https://app.daily.dev/posts/XbGrpoJDs" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AI agent security risks are no joke</strong>: AI agents are facing significant security risks, with reports of a BBC reporter hacked via an AI platform and critical vulnerabilities found in open-source agents like OpenClaw, highlighting the need for new security models and governance. <a href="https://app.daily.dev/feed-by-ids?id=1HpSP7L9S&amp;id=1TelNKsBD&amp;id=9xLrRFK2b&amp;id=FuKqPJxtz&amp;id=Jm3weKMWA&amp;id=ZmcjIu8ai&amp;id=em7afQ5sm&amp;id=kdJWfknAe&amp;id=ohhVPexbM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TypeScript 6 paves the way for a go rewrite</strong>: TypeScript 6 has been released as a transitional update, primarily to prepare for a planned rewrite of the compiler in Go, aiming to address performance bottlenecks in large-scale applications. <a href="https://app.daily.dev/posts/tiMDdyFcG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AWS enables event-driven architectures for RDS SQL server</strong>: AWS has enabled Lambda function triggers directly from RDS for SQL Server database events, allowing for event-driven architectures that decouple business logic from the database. <a href="https://app.daily.dev/posts/KzhIuYzeh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NTransformer speeds up local LLM inference</strong>: NTransformer, a C++/CUDA LLM inference engine, claims an 83x speedup for Llama 3.1 70B on a single RTX 3090, showcasing advancements in efficient local LLM inference. <a href="https://app.daily.dev/feed-by-ids?id=Ahx4zgQks&amp;id=DyVC0qyKG&amp;id=Fkac7nXWM&amp;id=K01E44V9h&amp;id=WoJMuyqAQ&amp;id=mulSXvfKt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LibreOffice calls out OnlyOffice as ‘fake open source’</strong>: The Document Foundation, stewards of LibreOffice, accused OnlyOffice of being ‘fake open source’ for allegedly partnering with Microsoft to enforce vendor lock-in through proprietary OOXML file formats. <a href="https://app.daily.dev/feed-by-ids?id=840Dw8MXb&amp;id=OwmBaq9Nm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloud skills still matter, but with an AI twist</strong>: A former AWS solutions architect suggests that cloud skills remain valuable but now require AI knowledge and a strong personal brand, with all current AWS postings requiring on-site work. <a href="https://app.daily.dev/feed-by-ids?id=8yc6ZDNd6&amp;id=WAm3NKS9B&amp;id=djwfpUzVm&amp;id=gX2zbD1Ji" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux ecosystem updates</strong>: Linux 7.0 is seeing initial DRM kernel graphics driver fixes for AMDGPU and renewed patch activity for eCryptfs, while GNOME 50 beta and KDE Plasma 6.6 bring new features and clarifications on systemd dependencies. <a href="https://app.daily.dev/feed-by-ids?id=2U7gOjLpG&amp;id=BZn5Av50T&amp;id=LutMi5Kcz&amp;id=MGUcuQgst&amp;id=gNaBG2osm&amp;id=u9gGdIsNO" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Altman dismisses AI water usage claims, pivots to human inefficiency</strong>: OpenAI CEO Sam Altman dismissed claims of high AI water usage as “completely untrue,” while simultaneously acknowledging total energy consumption as a legitimate concern. He then tried to frame AI training costs against human development, claiming AI is likely more energy-efficient per inference query than a human answering the same question. Because, you know, humans are just really inefficient computers. <a href="https://app.daily.dev/feed-by-ids?id=95MvV75SN&amp;id=mGa3eARS3&amp;id=ogUHau6h0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Gemini 3.1 pro: benchmark king, real-world flop</strong>: Google’s Gemini 3.1 Pro supposedly achieved top benchmark scores, including 100% on a custom skateboarding spatial reasoning benchmark. Yet, despite this alleged benchmark dominance, it’s plagued by practical issues like unreliable tool calling, looping behavior, and a buggy CLI. Apparently, Google is still “benchmaxing” at the expense of real-world usability, because who needs a working product when you have impressive numbers? <a href="https://app.daily.dev/feed-by-ids?id=95MvV75SN&amp;id=mGa3eARS3&amp;id=ogUHau6h0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>A16z partner says ‘vibe coding’ is ‘flat wrong’</strong>: A16z general partner Anish Acharya argues that “vibe coding everything” is “flat wrong” because software is only 8-12% of company expenses, making AI rebuilding of internal tools low-ROI. He suggests companies should direct AI towards core business development and optimizing the other 90% of costs. Because, clearly, the only thing AI is good for is cutting costs, not, you know, innovating or anything. <a href="https://app.daily.dev/feed-by-ids?id=8yc6ZDNd6&amp;id=WAm3NKS9B&amp;id=djwfpUzVm&amp;id=gX2zbD1Ji" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
