<h2>Must know</h2>
<ul>
<li><strong>TypeScript 6.0 bridges the gap to a Go native future</strong>: TypeScript 6.0 has officially launched as the final JavaScript-based version, introducing support for the Temporal API and Map upsert methods while aggressively deprecating legacy features like CommonJS and AMD targets. This release serves as a stabilization layer for the upcoming TypeScript 7.0, which features a Go-native compiler that is already cutting build times by 50% in early benchmarks. Developers should note that strict mode and ES2025 targets are now the default, signaling a clean break from legacy web patterns. <a href="https://app.daily.dev/posts/kynZFJI6r" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google and Snyk pivot to AI agent security at RSAC</strong>: At RSA Conference 2026, Google launched a Gemini-powered dark web intelligence service capable of processing 10 million posts daily to identify autonomous threats with a claimed 98% accuracy. Simultaneously, Snyk introduced its Agent Security solution, which includes specialized agents for AI Bill of Materials mapping and real-time guardrails for tools like Claude Code and Cursor. These releases address a critical governance gap, as new data suggests 59% of enterprises now suspect ‘shadow AI’ agents are operating within their networks without oversight. <a href="https://app.daily.dev/feed-by-ids?id=40Ri2RlBf&amp;id=iRGLDmExe&amp;id=0roBYJAi3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Evaluation driven development replaces the training loop</strong>: As AI agents commoditize code generation, the engineering bottleneck is shifting from writing code to systematic verification and intent alignment. Experts from <a href="http://monday.com" target="_blank" rel="noopener nofollow">monday.com</a> and Anthropic argue that ‘Evaluation-Driven Development’—using golden datasets and LLM-as-judge systems—is the new feedback loop replacing traditional model training. This shift is urgent, as recent studies show developers using AI score 17% lower on conceptual understanding, risking long-term skill atrophy if they fail to treat agents as teammates rather than black boxes. <a href="https://app.daily.dev/feed-by-ids?id=seweaAUF7&amp;id=EQi7fPMBC&amp;id=wVRLKEI7Y" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AMD pghot subsystem targets Linux kernel 7.1</strong>: A new kernel subsystem for hot-page tracking aims to unify page promotion logic across NUMA nodes and CXL memory tiers. <a href="https://app.daily.dev/posts/9z2uCLEQw" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Liminal runs Laravel 11 entirely in the browser</strong>: Using WebAssembly and PHP 8.4, this IDE enables full Laravel development with SQLite and local folder sync via the File System Access API. <a href="https://app.daily.dev/posts/ER7oeWZFY" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NixOS emerges as the ultimate sandbox for AI agents</strong>: Developers are increasingly using Nix flakes to provide agents with deterministic, isolated environments that prevent host system mutation. <a href="https://app.daily.dev/posts/on18fcwiv" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LiteLLM and RouteLLM tackle runaway agent costs</strong>: New open-source gateways allow for hard spend caps per team and binary routing that sends 85% of queries to cheaper models. <a href="https://app.daily.dev/posts/WNCyok6uu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Haskell dataframe 1.0 brings compile time schema safety</strong>: The library now features a typed API that catches column name errors at compile time and supports Apache Arrow for Polars interoperability. <a href="https://app.daily.dev/posts/jPEN94yLz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Intel FRED enabled by default for Panther Lake</strong>: The Flexible Return and Event Delivery feature is finally being toggled on in Linux to reduce privilege-level transition latency. <a href="https://app.daily.dev/feed-by-ids?id=G3fsiNuau&amp;id=5l3MiWfx9" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Unikraft boots one million VMs in ten milliseconds</strong>: By leveraging Firecracker and compressed snapshots, this approach provides VM-grade isolation for AI agent sandboxes at container density. <a href="https://app.daily.dev/posts/ZIuVqXL43" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The 98 percent accuracy hallucination</strong>: Google is claiming its new Gemini-powered threat intelligence hits 98% accuracy, a figure that conveniently ignores the industry’s baseline of 80% false positives. It’s the same old ‘AI magic’ marketing that promises to solve security by throwing more tokens at the problem while ignoring the underlying data quality issues. If you believe a black-box agent can autonomously classify 10 million dark web posts without a human in the loop, I have some NFT-backed bridge real estate to sell you. <a href="https://app.daily.dev/feed-by-ids?id=40Ri2RlBf&amp;id=0roBYJAi3" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vibe coding is just technical debt with a better UI</strong>: The ‘vibe coding’ trend is reaching peak absurdity, with startups raising millions to help developers ‘prompt’ their way out of understanding their own architecture. As noted by senior engineers at RSAC, this ‘just fix it’ loop often wastes more time in debugging than manual coding ever did. We’re essentially rebranding ‘untested spaghetti code’ as a revolutionary new paradigm because it’s easier than teaching people how a compiler works. <a href="https://app.daily.dev/feed-by-ids?id=EQi7fPMBC&amp;id=86Tfoe35V" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The subsidized 200 dollar productivity myth</strong>: Anthropic’s Claude Code plan costs $200 a month, but insiders suggest the actual compute cost is closer to $5,000. This isn’t a sustainable business model; it’s a VC-funded customer acquisition play designed to make you dependent on an agent that will inevitably get a 20x price hike or a massive lobotomy once the burn rate matters. Enjoy the ‘free’ compute while it lasts, but don’t build your entire workflow on a subsidy that’s one board meeting away from vanishing. <a href="https://app.daily.dev/feed-by-ids?id=fwselnkdq&amp;id=xeVJ3OUFc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
