<h2>Must know</h2>
<ul>
<li><strong>AI agents face critical security flaws</strong>: Recent incidents highlight severe security vulnerabilities in AI agents, with a BBC reporter hacked via an AI coding platform and OpenClaw, a popular open-source agent, revealing 512 vulnerabilities including remote code execution. These issues stem from agents’ access to private data, untrusted content ingestion, and external communication, creating a ‘lethal trifecta’ that traditional security models can’t handle, as reported by multiple sources. <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 prepares for Go rewrite</strong>: TypeScript 6 has been released as a transitional version, primarily to pave the way for a complete rewrite of the compiler in Go, aiming to fix significant performance bottlenecks in large applications. Developers are strongly encouraged to upgrade and resolve deprecation warnings, as this release enables strict mode by default and updates module resolution to ES modules, according to Bruno Couriol and other reports. <a href="https://app.daily.dev/posts/tiMDdyFcG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI introduces harness engineering</strong>: OpenAI has unveiled ‘Harness Engineering,’ an internal AI-driven methodology that uses Codex agents to automate software development tasks like code generation, testing, and CI workflows. A five-month experiment reportedly produced a million lines of code without manual writing, shifting engineers’ roles to designing environments and providing feedback, a concept Martin Fowler noted as a valuable framing for AI-enabled development. <a href="https://app.daily.dev/posts/dPCNsAxds" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Anthropic and Figma integrate Claude Code for design workflow</strong>: Anthropic and Figma have partnered to allow Claude Code to convert production code into editable Figma designs and round-trip updates back to code, streamlining designer-developer collaboration. <a href="https://app.daily.dev/posts/aCaE5xJ9z" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare R2 launches local uploads for faster writes</strong>: Cloudflare’s R2 object storage now offers ‘Local Uploads’ in open beta, reducing cross-region write latency by up to 75% by writing data to the nearest PoP and asynchronously replicating it to the home region. <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 functions to be triggered directly from RDS for SQL Server database events, enabling event-driven architectures by parsing SQL Server’s ERRORLOG for specific messages. <a href="https://app.daily.dev/posts/KzhIuYzeh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Local LLM inference gets a boost with NTransformer</strong>: NTransformer, a new 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, while Cord offers a new Python framework for orchestrating AI agents. <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>.NET 11 Preview 1 released with async improvements and C# 15</strong>: .NET 11 Preview 1 introduces runtime-level async improvements, Zstandard compression support, and new C# 15 language features, signaling the direction for the next major .NET release. <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>Linux ecosystem sees kernel fixes and desktop environment updates</strong>: Linux 7.0 is getting DRM kernel graphics driver fixes for AMDGPU and renewed patch activity for eCryptfs, while GNOME 50 beta includes variable refresh rate and fractional scaling, and KDE Plasma 6.6 released with a new login manager. <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 privacy-focused Android alternative</strong>: GrapheneOS provides a privacy-first Android experience by removing Google Play Services by default, offering sandboxed Google Play as an option, and including features like duress PIN and storage scopes. <a href="https://app.daily.dev/feed-by-ids?id=840Dw8MXb&amp;id=OwmBaq9Nm" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>OpenAI dismisses AI water usage concerns</strong>: Sam Altman, CEO of OpenAI, is out here claiming that reports of high AI water usage are ‘completely untrue,’ specifically refuting the 17 gallons per ChatGPT query figure. He then pivots to ‘human development takes 20 years and enormous resources’ as if that’s a valid comparison. Sure, Sam, because humans are powered by server farms and require liquid nitrogen cooling. IEEE Spectrum, meanwhile, reports AI data centers are exploring high-temperature superconductors because power constraints are so severe. Who are we supposed to believe? <a href="https://app.daily.dev/feed-by-ids?id=95MvV75SN&amp;id=jbBqj45d9&amp;id=ogUHau6h0" 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 sounding the alarm that AI startups building ‘LLM wrappers with thin IP’ and ‘AI aggregators’ are basically doomed, comparing them to early AWS resellers who got squeezed out. He’s bullish on ‘vibe coding platforms,’ though. So, if your AI startup isn’t doing something truly groundbreaking, or if it’s just making existing models slightly easier to use, Google thinks you’re toast. Thanks for the heads-up, Captain Obvious, TechCrunch. <a href="https://app.daily.dev/posts/djwfpUzVm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-powered design tool ‘Inception’ omits controls for chat interface</strong>: A new AI-powered design tool called ‘Inception’ is ditching traditional design controls for a ‘simple chat interface’ layered on a multi-frame canvas, with the LLM ‘natively thinking in Tailwind.’ Because who needs precise control or a visual hierarchy when you can just chat with your design tool? It’s designed for ‘rapid ideation,’ which I’m sure means rapid generation of generic, LLM-hallucinated UIs. <a href="http://Devtools.fm" target="_blank" rel="noopener nofollow">Devtools.fm</a> reports on this groundbreaking innovation. <a href="https://app.daily.dev/posts/g9hioZwpx" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
