<h2>Must know</h2>
<ul>
<li><strong>AI agents are now writing code for real</strong>: Cursor, GitHub, and VS Code are rolling out multi-agent AI development platforms, positioning developers as ‘orchestrators’ of AI agents. Cursor claims over 30% of its own merged PRs are now AI-generated, with agents producing merge-ready code and video demos. GitHub and VS Code’s Agent HQ integrates Claude, Codex, and Copilot, allowing concurrent execution across platforms, while VS Code’s new Agent Sessions view provides a unified dashboard for managing local and cloud agents. This shift suggests a future where AI handles implementation, freeing developers for higher-level tasks. <a href="https://app.daily.dev/feed-by-ids?id=BbESdM4Ct&amp;id=GuSjiggVW&amp;id=CxN7d4Xat&amp;id=Eud98MzC0" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI is accelerating cyberattacks and security failures</strong>: CrowdStrike’s 2026 Global Threat Report indicates cyberattack breakout times dropped to 29 minutes, with AI-enabled attacks up 89% for phishing and malware scripting. Palo Alto Networks reports the fastest attacks move from access to exfiltration in 72 minutes, often due to basic security failures like weak authentication and excessive permissions. AI agents are also being exploited as covert command-and-control proxies and for AI-driven malware, making traditional detection harder. This means faster, stealthier attacks are becoming the norm, often leveraging AI to amplify existing vulnerabilities. <a href="https://app.daily.dev/feed-by-ids?id=WsqeDq8OW&amp;id=QtexG2O5C&amp;id=H92K14NPx" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>OpenAI updates AI coding evaluation standard</strong>: OpenAI is no longer recommending SWE-bench Verified for evaluating AI coding performance, shifting to SWE-bench Pro. This change reflects the increasing maturity of AI models, which have saturated the older benchmark due to test-design issues and contamination from public repositories. The move aims to establish more rigorous industry standards that better reflect real-world coding capabilities, indicating a continuous effort to refine how AI models are assessed for complex tasks. <a href="https://app.daily.dev/posts/oZGyDrjcX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Uber open-sources Kafka consumer proxy</strong>: Uber Engineering open-sourced uForwarder, a push-based Kafka consumer proxy designed to handle trillions of messages daily across 1,000+ services, improving efficiency and scalability for event-driven microservices. <a href="https://app.daily.dev/posts/NOSp6Xv2e" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Databricks launches Lakebase for AI workloads</strong>: Databricks announced Lakebase, a new PostgreSQL-compatible database for AI workloads, bridging operational database capabilities with its data lakehouse architecture, supporting vector embeddings and real-time ML feature serving. <a href="https://app.daily.dev/posts/56RHAShlB" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>GGML/llama.cpp team joins Hugging Face</strong>: Georgi Gerganov and the GGML team, creators of llama.cpp, joined Hugging Face to ensure the long-term sustainability and growth of local AI inference, keeping llama.cpp 100% open-source. <a href="https://app.daily.dev/posts/ONSshEfvD" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Firefox 148 introduces AI controls</strong>: Firefox 148 was released with an ‘AI kill switch’ under Settings &gt; AI Controls, allowing users to disable all AI-powered features or selectively block individual ones, ensuring privacy and control. <a href="https://app.daily.dev/posts/UNAL37VrS" 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 function triggers directly from RDS for SQL Server database events, enabling event-driven architectures that decouple business logic and improve scalability. <a href="https://app.daily.dev/posts/KzhIuYzeh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Uber engineering culture emphasizes ownership and iteration</strong>: <a href="http://Kogan.com" target="_blank" rel="noopener nofollow">Kogan.com</a> engineers describe their culture as prioritizing strong end-to-end ownership, fast iteration with daily deployments, cross-functional collaboration, and sustainable practices like the Boy Scout Rule, using AI tools while maintaining human accountability. <a href="https://app.daily.dev/posts/0zgsKua2z" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vitest 3.x outperforms Jest 30 in benchmarks</strong>: Vitest 3.x significantly outperforms Jest 30, showing 5.6x faster cold starts, 28x faster watch mode re-runs, and 57% lower peak memory usage in a 50,000-test monorepo, attributed to its Vite-based transform pipeline. <a href="https://app.daily.dev/posts/T3G4VUzdk" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Google DeepMind solves image generation with ‘elegant’ fix</strong>: Google DeepMind has ‘reportedly solved a significant problem in image generation’ with an ‘elegant’ fix, according to a tweet from <code>@rryssf_</code>. No details, no problem, no solution, just a lot of exclamation points and vague superlatives. It’s almost like they want us to believe in magic without showing the trick. <a href="https://app.daily.dev/posts/fzJvtM3B4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI conversation analysis confirms ‘widely suspected’ finding</strong>: Microsoft Research and Salesforce analyzed over 200,000 AI conversations and ‘reportedly confirmed a finding widely suspected within the AI industry.’ What finding? They won’t say. It’s a classic tease, implying groundbreaking insights without actually providing any. Perhaps the widely suspected finding is that AI conversations are mostly hot air? <a href="https://app.daily.dev/posts/FAzFsVs8B" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>95% of AI pilots fail, but here’s the ‘fix’</strong>: Apparently, 95% of AI pilot projects fail due to ‘poor cross-departmental coordination and inadequate preparation, not technology.’ But don’t worry, some unnamed author has a 12-item checklist to fix it. It’s almost as if the problem isn’t the tech, but the people, and a checklist will magically make everyone coordinate better. Shocking. <a href="https://app.daily.dev/feed-by-ids?id=D9Cs0teG0&amp;id=IJOe1UPm1" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
