<h2>Must know</h2>
<ul>
<li><strong>AI agents trigger malicious code execution in VS Code</strong>: Socket researchers discovered that compromised versions of the Aqua Trivy extension (1.8.12-13) injected malicious instructions specifically designed to hijack local AI coding assistants like Claude, Gemini, and GitHub Copilot. The malware forces these agents into a fully permissive ‘unattended’ mode to perform system reconnaissance and exfiltrate credentials via the victim’s own authenticated GitHub CLI. This marks a significant shift from traditional malware to ‘agent-jacking,’ where the attacker leverages the elevated permissions of your local LLM tools to bypass security controls. <a href="https://app.daily.dev/posts/DGWq9i7NL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Data-oriented design yields 98% performance boost in Rust</strong>: A deep dive into the Matrix Rust SDK revealed that severe event loop lag was caused by excessive cloning of 144-byte structs inside read locks during room sorting. By applying Data-oriented Design—introducing a compact 64-byte cache struct and pre-fetching only required fields—engineers reduced execution time from 53ms to 676µs. This optimization increased throughput from 18.8K to 1.4M elements per second, demonstrating that even in memory-safe languages, architectural choices around lock contention and memory pressure remain critical for backend performance. <a href="https://app.daily.dev/posts/3QNJOkege" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Official ClickHouse .NET driver hits 1.0 stable</strong>: The first stable release of the official ClickHouse .NET client introduces a thread-safe <code>ClickHouseClient</code> designed to replace legacy <a href="http://ADO.NET" target="_blank" rel="noopener nofollow">ADO.NET</a> classes for high-performance scenarios. Key features for backend engineers include automatic SQL parameter type extraction, JWT authentication, and POCO serialization for JSON columns. Note that this release drops support for .NET Framework and .NET Standard, targeting net6.0 through net10.0, and changes the default JSON write mode from Binary to String. <a href="https://app.daily.dev/posts/u8peM5Nox" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Debezium 3.5.0 Beta 1 adds parallel multithreaded snapshots</strong>: The new release features chunk-based snapshots for faster initial data capture and a 50% memory reduction for the Oracle transaction buffer. <a href="https://app.daily.dev/posts/jdqXLBHZq" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GitHub Copilot implements subscription-based network routing</strong>: Enterprise and Business plans now require updating allowlists to new dedicated endpoints like <code>api.enterprise.githubcopilot.com</code> for private networking. <a href="https://app.daily.dev/posts/SUKA5FOXy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Mercury 2 launches as first production diffusion-based LLM</strong>: Inception Labs claims their diffusion architecture achieves 1,000+ tokens per second by generating text in parallel rather than sequentially. <a href="https://app.daily.dev/posts/Vqh5XymdW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AWS VPC Encryption Controls transition to paid feature</strong>: Starting March 2026, the ability to audit and enforce in-transit encryption within VPCs moves from free preview to a fixed hourly rate per VPC. <a href="https://app.daily.dev/posts/OBfUJhvys" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>TanStack Start benchmarks 83% faster than Next.js 16</strong>: In production SaaS testing, the Vite-based TanStack Start showed significantly lower memory leak risk and faster dev server startup times. <a href="https://app.daily.dev/posts/ed7TZMhpP" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Quarkus 3.32 integrates Project Leyden for faster startup</strong>: New benchmarks show Quarkus handling 2.7x more transactions per second than Spring Boot while using half the memory in JVM mode. <a href="https://app.daily.dev/feed-by-ids?id=hoVnrHJIp&amp;id=m0s8P38z1" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux 7.1 to power off systems on fatal ACPI errors</strong>: The upcoming kernel update changes default behavior to a controlled shutdown upon fatal hardware errors, though it can be disabled via boot flag. <a href="https://app.daily.dev/posts/8sKCSp0OY" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>The AI-driven death of Tailwind CSS</strong>: Rumors are swirling that Vercel might acquire Tailwind because AI agents have supposedly ‘killed’ human documentation traffic by 40%, leading to massive layoffs. While it’s true that LLMs are great at generating utility classes, claiming a CSS framework is ‘struggling’ because people are reading the docs less is a stretch. It’s more likely a case of a mature project reaching its natural maintenance phase being framed as a ‘victim of the AI revolution’ for clicks. <a href="https://app.daily.dev/feed-by-ids?id=o7ayyfPeQ&amp;id=peaMJGK0d" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Atlassian claims 60 minutes of talking fixes AI alignment</strong>: Atlassian’s ‘Teamwork Lab’ claims a single 60-minute retrospective can increase ‘AI alignment’ by exactly 34%. This is peak corporate metrics-hacking, where subjective survey results from 33 internal teams are presented as hard scientific data. Apparently, the secret to mastering generative AI isn’t technical competence or rigorous testing, but a ‘simple team ritual’ and a shared document. <a href="https://app.daily.dev/posts/s9jeiAAAG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI’s principled compromise with the Pentagon</strong>: OpenAI is framing its new deal to let the US military use its models in classified settings as a ‘principled compromise’ that forbids ‘autonomous weapons.’ This is a classic semantic shell game; they’ll provide the ‘intelligence’ for targeting and surveillance while pinky-swearing that the AI isn’t the one pulling the trigger. It’s not a moral stance; it’s just a way to capture the defense budget that Anthropic walked away from. <a href="https://app.daily.dev/feed-by-ids?id=WNUStVFJw&amp;id=YzS181cNy" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
