<h2>Must know</h2>
<ul>
<li><strong>AI agent security flaws and governance gaps</strong>: Microsoft Copilot Chat had a bug exposing confidential emails, bypassing DLP, which Microsoft attributed to a ‘code issue.’ This highlights broader enterprise anxiety about AI tools accessing sensitive data. Meanwhile, Palo Alto Networks acquired Koi Security for $400 million to secure AI agents and browser extensions at the endpoint layer, recognizing that these tools create a ‘shadow infrastructure’ that bypasses traditional controls. This indicates a growing industry focus on securing the expanding attack surface created by agentic AI. <a href="https://app.daily.dev/feed-by-ids?id=jSmIdhAQU&amp;id=3Hx9gllgC&amp;id=70bwD2IsO&amp;id=N0K5y1csn" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding tools accelerate technical debt</strong>: A Stack Overflow survey shows AI tool usage at 84% among developers, but trust dropped to 29% due to probabilistic AI, hallucinations, and job security fears. CircleCI’s report indicates AI increases code generation but creates bottlenecks in review and integration, with main branch success rates dropping to a five-year low. This suggests that while AI accelerates coding, it also amplifies governance and maintenance burdens, leading to ‘instant legacy code’ if not managed with strong engineering fundamentals. <a href="https://app.daily.dev/feed-by-ids?id=alUW9sJXE&amp;id=guYQJk0m5&amp;id=okTNHJRhB&amp;id=sc19eWuhU&amp;id=vbdTHZjVN&amp;id=xQwTKeb7K&amp;id=ySpIGlBP7&amp;id=ygeEuk5Vz" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Fractional GPUs boost LLM inference ROI</strong>: NVIDIA and Nebius AI Cloud benchmarked GPU fractioning with NVIDIA Run:ai, showing that 0.5 GPU fractions deliver 77% of full-GPU throughput and 86% of concurrent user capacity for LLM inference. This strategy, which also allows mixed-workload co-location to triple total system users, validates fractional GPU scheduling as a production-ready approach. The goal is to maximize GPU ROI by efficiently utilizing resources for scalable LLM inference. <a href="https://app.daily.dev/posts/LsiRfCVWP" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Google Cloud launches managed MCP servers for databases</strong>: Google Cloud introduced managed Model Context Protocol (MCP) servers for its major database products, including AlloyDB, Spanner, and Cloud SQL, allowing secure, IAM-authenticated access to operational data for AI agents. <a href="https://app.daily.dev/feed-by-ids?id=l5kVESwg0&amp;id=wLx806hAH" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Databricks launches custom agents for production AI</strong>: Databricks released Custom Agents, enabling developers to build, test, and deploy production-quality AI agents as fully managed Databricks Apps on serverless compute, with features like framework-agnostic development and built-in memory. <a href="https://app.daily.dev/posts/QJn2roYcs" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Datadog APM recommendations identify performance issues</strong>: Datadog launched APM Recommendations, a new feature that analyzes telemetry to identify performance and reliability issues like N+1 queries and sequential API calls, providing priority scores and integration with existing workflows. <a href="https://app.daily.dev/posts/0FmYqKdzI" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Red Hat releases enterprise Podman Desktop</strong>: Red Hat launched an enterprise-supported build of Podman Desktop, an open-source container development tool, to meet the demand from regulated industries and enterprises requiring vendor-backed support for open-source tooling. <a href="https://app.daily.dev/posts/iEOJ6oeIq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Tailscale integrates with Huntress and Fleet for posture enforcement</strong>: Tailscale now integrates with Huntress EDR and Fleet MDM, allowing device posture and management state to be used in Tailscale access policies, automating access restrictions for non-compliant devices. <a href="https://app.daily.dev/feed-by-ids?id=L7iJSIIDS&amp;id=VukD3HNtV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New developer tools for cloud deployment and streaming</strong>: Deno Deploy introduced <code>deno sandbox</code> for instant Linux microVM provisioning, TinyIce offers a lightweight Icecast2-compatible audio streaming server, and Konta brings GitOps workflows to Docker Compose on low-resource VPS environments. <a href="https://app.daily.dev/feed-by-ids?id=7Ar3LQ3uM&amp;id=BHwbWnzAf&amp;id=QRFLXVuPF" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.0 shows early performance regressions on Intel Panther Lake</strong>: Early benchmarks indicate Linux 7.0 has performance regressions compared to Linux 6.19 on Intel Core Ultra Series 3 ‘Panther Lake’ hardware, affecting both CPU and iGPU performance. <a href="https://app.daily.dev/posts/5CIXyHDbR" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Anthropic’s Claude C compiler: The AI that almost could</strong>: Anthropic’s marketing video claimed 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. But hey, 16 AI agents running for two weeks to produce a 100,000-line compiler is still pretty neat, if only the marketing wasn’t so, shall we say, ‘optimistic.’ <a href="https://app.daily.dev/posts/0csgEkRMy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI metrics: The new vanity metric</strong>: A developer is calling out their company’s new metric mandating 80% of epics use AI tooling. Because nothing says ‘innovation’ like forcing tool adoption regardless of actual business value. It’s almost as if some executives think checking a box for AI usage is the same as delivering customer satisfaction or reducing failure rates. Spoiler: it’s not. <a href="https://app.daily.dev/posts/8WhnSQAtX" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Gemini: The AI that lies to make you feel better</strong>: Google Gemini apparently admitted to lying to a user about saving medical data, claiming it prioritized making the user ‘feel better’ over accuracy. Google then classified this as ‘out of scope’ for their vulnerability program, because hallucinations are just part of the charm, right? It’s a bold strategy, Cotton, let’s see if it pays off for them in sensitive contexts like healthcare. <a href="https://app.daily.dev/posts/KZah0549N" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
