<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, testing, and CI workflows. 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 bridge code and design with claude</strong>: Anthropic and Figma have teamed up to integrate Claude Code, allowing developers to convert production code into editable Figma designs. This workflow enables sending live pages to Figma files with a simple prompt, facilitating design collaboration and round-tripping updates back to the codebase. The integration aims to streamline the often-clunky designer-developer handoff. <a href="https://app.daily.dev/posts/aCaE5xJ9z" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TypeScript 6 prepares for a go rewrite</strong>: TypeScript 6 has been released, primarily as a transitional step to prepare for a planned rewrite of the TypeScript compiler in Go, aiming to solve performance bottlenecks. Developers are encouraged to upgrade to TypeScript 6 to get ready for TypeScript 7, which will feature the Go rewrite. Key changes include strict mode by default, ES modules, and deprecation of older targets. <a href="https://app.daily.dev/posts/tiMDdyFcG" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Cloudflare R2 speeds up uploads with local writes</strong>: AWS launched “Local Uploads” for R2 object storage, reducing cross-region write latency by up to 75% by writing data to the nearest PoP and asynchronously replicating it. <a href="https://app.daily.dev/posts/XbGrpoJDs" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AWS enables event-driven architectures for RDS SQL server</strong>: AWS now allows Lambda function triggers directly from RDS for SQL Server database events, enabling event-driven architectures using stored procedures and CloudWatch Logs. <a href="https://app.daily.dev/posts/KzhIuYzeh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>.NET 11 preview 1 brings async improvements and C# 15</strong>: .NET 11 Preview 1 introduces runtime-level async improvements, Zstandard compression support, and new C# 15 language features. <a href="https://app.daily.dev/feed-by-ids?id=QBAQQNpLN&amp;id=xPZZ4nMO0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NTransformer boosts local LLM inference speed</strong>: NTransformer, a C++/CUDA LLM inference engine, claims an 83x speedup for Llama 3.1 70B on a single RTX 3090, using adaptive caching and layer skipping. <a href="https://app.daily.dev/posts/Ahx4zgQks" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cord framework orchestrates AI agents</strong>: Cord is a new open-source Python framework for orchestrating AI agents, allowing a root Claude agent to dynamically decompose tasks and spawn or fork other agents. <a href="https://app.daily.dev/posts/DyVC0qyKG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux ecosystem sees kernel and desktop environment updates</strong>: Linux 7.0 is getting 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. <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>
<li><strong>GrapheneOS offers a privacy-centric android experience</strong>: GrapheneOS continues to be a privacy-focused Android alternative, offering a de-Googled experience with sandboxed Google Play as an option for necessary apps. <a href="https://app.daily.dev/posts/840Dw8MXb" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Altman dismisses AI water usage claims, compares AI to raising a child</strong>: OpenAI CEO Sam Altman is out here dismissing claims of high AI water usage as “completely untrue,” while simultaneously acknowledging total energy consumption is a legitimate concern. He then tries to frame AI training costs against human development, because apparently, comparing a data center to raising a child is a totally valid metric. Sure, Sam, whatever helps you sleep at night. <a href="https://app.daily.dev/feed-by-ids?id=95MvV75SN&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 is apparently a benchmark champion, scoring 78% on ARC AGI 2 and 100% on a skateboarding spatial reasoning test. Yet, in the real world, it’s plagued by unreliable tool calling, looping, hallucinated packages, and a buggy CLI. So, Google is “benchmaxing” at the expense of actual usability. Because who needs a functional AI when you have impressive, meaningless numbers? <a href="https://app.daily.dev/posts/mGa3eARS3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google VP warns against ‘thin ip’ AI startups</strong>: Google Cloud VP Darren Mowry is warning that AI startups focused on LLM wrappers with “thin IP” and AI aggregators are basically doomed. He’s drawing parallels to early AWS resellers, who, let’s be honest, were just reselling. So, if your brilliant AI idea is just putting a fancy UI on someone else’s model, prepare for margin pressure. Apparently, actual innovation is still a thing. <a href="https://app.daily.dev/posts/djwfpUzVm" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
