<h2>Must know</h2>
<ul>
<li><strong>AI agent security is the new frontier</strong>: The proliferation of AI agents is creating a massive new attack surface, with machine identities now outnumbering human identities by 82:1 in enterprise environments. Traditional security controls are failing, leading to new solutions like StrongDM ID’s identity provider for AI agents and Palo Alto Networks acquiring Koi Security for endpoint protection of these autonomous tools. This shift demands a focus on verifiable identities, human sponsorship, and pre-deployment scanning for vulnerabilities in AI agent configurations, as highlighted by <a href="http://Mend.io" target="_blank" rel="noopener nofollow">Mend.io</a>. <a href="https://app.daily.dev/feed-by-ids?id=U1HM18ytf&amp;id=JQP8vhpTo&amp;id=PKHQykURz&amp;id=3Hx9gllgC&amp;id=70bwD2IsO&amp;id=N0K5y1csn&amp;id=sX4nQIjj6&amp;id=v09JiOK4R&amp;id=hDN0CJdoX&amp;id=sX4nQIjj6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s sonnet 4.6 is a token muncher</strong>: Anthropic launched Claude Sonnet 4.6, touting it as 40% cheaper than Opus 4.6 with improved benchmarks. However, independent analysis by Artificial Analysis reveals that Sonnet 4.6 uses over four times more tokens for complex tasks due to its ‘adaptive thinking’ approach. This ‘token muncher’ problem means that while the per-token price might be lower, the actual cost for heavy, extended-thinking use cases could negate any savings, making its cost-effectiveness questionable in practice. <a href="https://app.daily.dev/feed-by-ids?id=i4O6sQBIw&amp;id=j12nF8mSg&amp;id=jj2Kauaop&amp;id=lUz1VvbAr&amp;id=y0N7mzotz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agent development is getting real</strong>: Databricks launched Custom Agents, enabling developers to build, test, and deploy production-quality AI agents as fully managed serverless apps. This move, along with Google Cloud’s new managed Model Context Protocol (MCP) servers for its databases, signifies a major push towards robust, enterprise-grade infrastructure for agentic AI. These platforms aim to accelerate the journey from prototype to production, offering framework-agnostic development, CI/CD integration, and secure access to enterprise data for AI agents. <a href="https://app.daily.dev/feed-by-ids?id=QJn2roYcs&amp;id=l5kVESwg0&amp;id=wLx806hAH&amp;id=4idREqj8d&amp;id=CoCDQAhlH&amp;id=J0baIl5WX&amp;id=s4jNANjno&amp;id=R2tIcO414&amp;id=D45i7bkfM&amp;id=EZOdc00Xc" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Google launches lyria 3 AI music generation</strong>: Google integrated Lyria 3, an AI music generation model, into Gemini, allowing users to create 30-second songs with lyrics from text, photos, or videos, complete with SynthID watermarking for AI identification. <a href="https://app.daily.dev/posts/eN6VW0Hvt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA run:ai boosts GPU efficiency for LLMs</strong>: NVIDIA Run:ai’s GPU fractioning technology allows 0.5 GPU fractions to deliver 77% of full-GPU throughput and 86% of concurrent user capacity for LLM inference, significantly maximizing GPU ROI. <a href="https://app.daily.dev/posts/LsiRfCVWP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google chrome zero-day under active exploitation</strong>: Google released an urgent update for Chrome to patch CVE-2026-2441, a high-severity use-after-free zero-day vulnerability in CSS font feature parsing that is actively being exploited. <a href="https://app.daily.dev/posts/ISqaRw1RF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Critical vulnerabilities in popular VS code extensions</strong>: OX Security disclosed critical vulnerabilities in popular VS Code extensions like Live Server and Code Runner, affecting 128 million installs and enabling file theft, RCE, and local network reconnaissance. <a href="https://app.daily.dev/posts/IuDICegHN" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vercel speeds up webstreams 10x</strong>: Vercel developed <code>fast-webstreams</code>, a drop-in replacement for WHATWG WebStreams that achieves 10-14x throughput gains by optimizing Node.js’s internal <code>pipeline()</code> for server-side rendering in Next.js. <a href="https://app.daily.dev/posts/OTXPVGIi5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Tailscale peer relays are generally available</strong>: Tailscale announced the general availability of Peer Relays, allowing users to deploy high-throughput relay nodes on any Tailscale device to improve network connectivity and performance, even in restrictive cloud environments. <a href="https://app.daily.dev/posts/Q0uzlfbeJ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Red hat launches enterprise podman desktop</strong>: Red Hat released an enterprise-supported version of Podman Desktop, providing vendor-backed support for the open-source container development tool to meet the demands of regulated industries and risk-averse enterprises. <a href="https://app.daily.dev/posts/iEOJ6oeIq" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI productivity gains are still a myth for most</strong>: A survey of nearly 6,000 executives found that over 80% reported no measurable impact from AI on productivity or employment over the past three years. Despite 69% of businesses using some form of AI, the anticipated 1.4% productivity gains and 1.75 million job eliminations by 2028 remain largely theoretical. So, while the hype machine keeps churning, most companies are still waiting for AI to actually do something useful, beyond generating more PowerPoint slides, I guess. <a href="https://app.daily.dev/feed-by-ids?id=rfmYnNc4e&amp;id=FtPdxfWs5&amp;id=GcBedQWU0&amp;id=XYiUFiLYo" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Anthropic’s C compiler: a triumph of marketing over reality</strong>: Anthropic released a marketing video claiming Claude built a C compiler ‘from scratch’ with ‘no human intervention’ that could compile Linux. Turns out, it used 37 years of GCC test suites, an online GCC oracle, required human restarts, and cost $20,000 in API calls. Oh, and it couldn’t actually boot Linux. It’s a genuine achievement for 16 AI agents to autonomously produce a 100,000-line compiler, but why ruin a good story with, you know, facts? <a href="https://app.daily.dev/posts/0csgEkRMy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The pentagon vs anthropic: a very normal dispute</strong>: The Pentagon is threatening to label Anthropic a national security risk because Claude AI was allegedly used in a military operation in Venezuela, and Anthropic dared to ask about it. The DoD interpreted this inquiry as a ‘condemnation’ of the mission, sparking a feud. Apparently, asking questions about your AI being used in a military operation that resulted in casualties is now considered an act of war. OpenAI, Google, and XAI, meanwhile, have agreed to ‘all lawful purposes’ for military use. Good to know where everyone stands on the ‘don’t be evil’ scale. <a href="https://app.daily.dev/feed-by-ids?id=9eGp6hlhZ&amp;id=3Lrs6Yolt&amp;id=3NTLSs4Mu&amp;id=dysAuzdpU&amp;id=hDN0CJdoX&amp;id=sX4nQIjj6&amp;id=v09JiOK4R" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
