<h2>Must know</h2>
<ul>
<li><strong>AI agent security is a dumpster fire</strong>: Multiple reports highlight catastrophic security flaws in AI agents. A BBC reporter was hacked via an AI coding platform, Orchards, due to a vulnerability allowing malicious code injection. OpenClaw, a popular open-source agent, revealed 512 vulnerabilities, including critical remote code execution, with 12-20% of its marketplace packages being malicious. This ‘lethal trifecta’ of data access, untrusted content, and external communication means traditional security models are completely inadequate, as detailed by sources like WIRED and the BBC. <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>OpenAI introduces harness engineering for AI-driven development</strong>: OpenAI has unveiled ‘Harness Engineering,’ an internal methodology using Codex agents to automate large-scale 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. Martin Fowler noted this as a significant framing of AI-enabled software development, suggesting a future where AI agents handle much of the coding grunt work. <a href="https://app.daily.dev/posts/dPCNsAxds" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TypeScript 6 prepares for a go rewrite</strong>: TypeScript 6 has been released as a transitional version, primarily to prepare the ecosystem for a planned rewrite of the compiler in Go, aiming to solve performance bottlenecks. This update enables strict mode by default, switches to ES modules, and deprecates older targets like ES5. Developers are strongly encouraged to upgrade now to ensure compatibility with the upcoming TypeScript 7, which will feature the Go-based compiler. <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>Anthropic and Figma integrate claude code for design-to-code workflow</strong>: Anthropic and Figma collaborated to allow Claude Code to convert production code into editable Figma designs and round-trip updates back to code, streamlining designer-developer workflows. <a href="https://app.daily.dev/posts/aCaE5xJ9z" 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 via CloudWatch Logs. <a href="https://app.daily.dev/posts/KzhIuYzeh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Cloudflare r2 local uploads launch</strong>: Cloudflare 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 to the home region. <a href="https://app.daily.dev/posts/XbGrpoJDs" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Sandkasten offers self-hosted sandbox runtime for AI agents</strong>: Sandkasten provides a self-hosted sandbox runtime for AI agents, offering stateful Linux sandboxes with persistent shell sessions and file operations using native Linux primitives for isolation. <a href="https://app.daily.dev/posts/kdJWfknAe" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NTransformer boosts local llm inference on gpus</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, layer streaming, and self-speculative decoding. <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>Datastar simplifies modern web development with backend-driven hypermedia</strong>: Datastar is an 11 KB backend-driven hypermedia framework that unifies HTMX and Alpine.js functionality, using Server-Sent Events for real-time push updates and offering reactive signals and DOM morphing. <a href="https://app.daily.dev/posts/GKcflwWUw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Ghostty 1.3 adds scrollbar support</strong>: Ghostty 1.3.0, a terminal emulator, will introduce scrollbar support, a highly requested feature, along with a command palette session search and tab renaming on Linux. <a href="https://app.daily.dev/posts/AthQk05VJ" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>OpenAI ceo dismisses ai water usage concerns, because humans are more expensive</strong>: Sam Altman, OpenAI’s CEO, confidently declared claims of high AI water usage ‘completely untrue,’ while acknowledging energy consumption is a ‘legitimate concern.’ He then pivoted to comparing AI training costs against human development, stating training a human takes 20 years and ‘enormous resources.’ So, apparently, AI is more energy-efficient than a human answering a question, which is a relief, because who needs water when you have efficient algorithms? <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>Google vp warns against ‘thin ip’ ai startups, while google benchmaxes</strong>: Google Cloud VP Darren Mowry is out here warning that AI startups relying on ‘thin IP’ LLM wrappers or acting as aggregators are doomed, drawing parallels to early AWS resellers. Meanwhile, Google’s Gemini 3.1 Pro is ‘benchmaxing’ with top scores but is plagued by practical issues like unreliable tool calling and looping behavior. It’s almost like focusing on benchmarks over real-world usability might be a ‘thin IP’ strategy in itself, but hey, at least they’re not just wrapping other models, right? <a href="https://app.daily.dev/feed-by-ids?id=8yc6ZDNd6&amp;id=WAm3NKS9B&amp;id=djwfpUzVm" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Unity ceo says ai will make entire games, because human creativity is overrated</strong>: Unity’s CEO announced they’ll unveil a beta at GDC that lets developers ‘prompt full casual games into existence using natural language only.’ Because who needs actual game designers, artists, or programmers when you can just type ‘Flappy Bird, but with cats’? While it might be good for ‘rapid prototyping,’ the idea that AI will replace professional game developers for anything beyond the most basic, uninspired casual games is, frankly, a bit of a stretch. But hey, if it means more Flappy Cat clones, I’m sure the world is ready. <a href="https://app.daily.dev/posts/VkFal2xDM" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
