<h2>Must know</h2>
<ul>
<li><strong>AI agents are coming for your code, and your job</strong>: GitHub has released a multi-platform SDK for integrating Copilot’s agentic workflows into applications, supporting Python, TypeScript, Go, and .NET. This SDK enables orchestration, planning, tool invocation, and programmatic file edits, essentially allowing AI agents to write production code autonomously, as reported by GitHub. Meanwhile, OpenAI Codex is now natively integrated into JetBrains IDEs, powered by GPT-5.2-Codex, offering adjustable autonomy levels for coding tasks like refactoring and security vulnerability detection, according to JetBrains. This means AI is not just suggesting code, it’s actively taking over more complex development tasks, raising questions about the future role of human developers. <a href="https://app.daily.dev/feed-by-ids?id=uAvdGTErT&amp;id=GXXnGjqMp&amp;id=GuG7K2ryZ&amp;id=xBP8EMs83" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI models still struggle with real-world tasks</strong>: Mercor’s Apex-Agents benchmark, which assesses AI models on white-collar tasks from consulting, investment banking, and law, revealed that leading models like Gemini 3 Flash and GPT-5.2 achieved a maximum accuracy of only 24%. This benchmark requires AI to synthesize information from various sources and apply detailed domain knowledge, highlighting a significant gap in AI’s ability to reliably replace knowledge workers in complex, multi-tool tasks. The findings suggest that while AI agents are making rapid progress, they are not yet ready for high-stakes knowledge work, as reported by Mercor. <a href="https://app.daily.dev/posts/OsmrZ9UpM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Ireland considers expanding police surveillance with spyware</strong>: Ireland has proposed the Communications (Interception and Lawful Access) Bill, which would update its 1993 surveillance framework to enable law enforcement to deploy commercial spyware for intercepting encrypted digital communications. This move comes amid a rise in spyware use across Europe, with countries like Greece and Hungary experiencing documented abuses. While the bill promises judicial oversight and privacy safeguards, specifics on implementation or independent oversight are not detailed, raising concerns about potential privacy implications. <a href="https://app.daily.dev/posts/wO3SoInHG" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>TypeScript embraces runtime type stripping for faster server-side workflows</strong>: TypeScript is moving towards ‘no-build’ workflows for server-side development by introducing type stripping, which removes type annotations at runtime, preserving line numbers and eliminating the need for source maps. <a href="https://app.daily.dev/posts/gB9bktBA8" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux kernel vulnerability allows unbounded memory consumption</strong>: A security flaw in the Linux kernel’s Direct Rendering Manager (DRM) graphics driver could allow unprivileged users to trigger unbounded kernel memory consumption, leading to system-wide out-of-memory conditions. <a href="https://app.daily.dev/posts/9SIWTOgdA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Datadog unifies RUM and APM data with automatic retention filters</strong>: Datadog has launched cross-product retention filters that automatically retain Application Performance Monitoring (APM) traces linked to Real User Monitoring (RUM) sessions, aiming to unify frontend and backend observability data for improved troubleshooting. <a href="https://app.daily.dev/posts/fs5040Sq6" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Vercel Sandboxes now support filesystem snapshots</strong>: Vercel Sandboxes has introduced filesystem snapshot support, allowing users to capture the complete state of a running sandbox environment and launch new sandboxes from these saved snapshots, streamlining development workflows. <a href="https://app.daily.dev/posts/Uv3Zx9otG" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Feast integrates with PyTorch for enhanced feature management</strong>: Feast, an open-source feature store, has joined the PyTorch Ecosystem to provide a unified API for consistent feature management across model training and deployment, addressing training-serving skew in production AI systems. <a href="https://app.daily.dev/posts/dPG1hvjH4" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Pulumi introduces replacementTrigger for automated resource recreation</strong>: Pulumi v3.215.0 now includes a <code>replacementTrigger</code> option, enabling developers to force infrastructure-as-code resource recreation automatically when defined conditions or metadata values change, useful for tasks like secret cycling. <a href="https://app.daily.dev/posts/ay0AhT4GQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>ClickHouse launches managed Postgres service with native integration</strong>: ClickHouse has unveiled a new managed Postgres service that promises up to 10X faster performance for disk-bound workloads and offers native integration with ClickHouse through Change Data Capture (CDC) and the pg_clickhouse extension. <a href="https://app.daily.dev/posts/WHTgaLabX" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>Google’s AI Mode wants to read your Gmail and photos</strong>: Google’s new AI Mode, available to AI Pro and AI Ultra subscribers, will access your Gmail and Google Photos if you opt-in. They claim it’s for suggesting vacation itineraries and shopping items, but conveniently omit details on privacy risks or how to control your data once you’ve let the AI genie out of the bottle. Because who doesn’t want an AI sifting through their personal life for ‘suggestions’? <a href="https://app.daily.dev/posts/Kw6SK0Mqr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents are the new ‘cowboy sysadmins’</strong>: Apparently, AI agents are introducing ‘non-deterministic actions’ into IT operations, which is being compared to the ‘cowboy’ sysadmins of yore who made unreproducible changes to production. The article warns against letting AI directly modify live systems, suggesting they stick to generating playbooks and configurations. Because, you know, we definitely need more AI-generated YAML files to debug. <a href="https://app.daily.dev/posts/U8PYVjc73" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-driven QA analytics promises to solve all your problems</strong>: Testlio has launched LeoInsights, an AI-based QA analysis solution trained on 13 years of their testing data, promising to produce executive-level summaries and measure efficiency improvements. It even does sentiment analysis on app reviews! Because what executives really need is an AI to tell them how much money they’re saving on QA, rather than, say, actual bug-free software. And of course, no independent validation or critical review is provided. <a href="https://app.daily.dev/posts/xZI6Dgtwy" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
