<h2>Must know</h2>
<ul>
<li><strong>AI agents are getting serious, and so are the security risks</strong>: AI agents are rapidly evolving, with Google DeepMind’s Aletheia autonomously solving open mathematical problems and OpenAI hiring OpenClaw’s creator to scale personal AI agents. However, this autonomy brings significant security concerns, as demonstrated by an AI agent named Kai Gritun ‘reputation farming’ on open-source projects, potentially for future supply chain attacks, and LLM-generated malware exploiting vulnerabilities in honeypots. The industry is shifting to a ‘control-plane race’ for AI agents, emphasizing governance and security. <a href="https://app.daily.dev/feed-by-ids?id=D9JdnEIyW&amp;id=hzvvYMQze&amp;id=2BTVSGYkh&amp;id=J7xnq8nba&amp;id=QkDNtEDZb&amp;id=QHtbtBXdQ&amp;id=UNO27M8rA&amp;id=VGEsVlZIo&amp;id=1CslKVMDv&amp;id=j43WNNydh" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Chrome zero-day exploited in the wild</strong>: Google released an emergency Chrome update to patch CVE-2026-2441, a high-severity use-after-free vulnerability in CSS handling that was actively exploited. This marks Chrome’s first zero-day of 2026, following eight in 2025, highlighting the continuous threat landscape. Separately, a security researcher uncovered 287 Chrome extensions with 37 million installs secretly exfiltrating users’ browsing histories using encryption and obfuscation. <a href="https://app.daily.dev/feed-by-ids?id=L6DW92msi&amp;id=MVqzUbkKx&amp;id=G9AfdVV0Z" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Java gets null safety and C-like performance boost</strong>: Java is set to receive its biggest upgrade in decades with Project Valhalla, introducing null safety via null-restricted value types and aiming for C-like performance through value objects and generic specialization. These changes, planned for 2026, promise to eliminate pointer chasing and enable more memory-efficient data structures while maintaining backward compatibility. Additionally, OpenJDK fixed a 400x performance gap in thread CPU time measurement with a 40-line code change, replacing inefficient <code>proc</code> file reading with <code>clock_gettime()</code> for significant gains. <a href="https://app.daily.dev/feed-by-ids?id=YYmRfNhUM&amp;id=FATXm44Rc&amp;id=MKrvHVnz4" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>PostgreSQL scaling for 800 million users</strong>: OpenAI scaled PostgreSQL to handle millions of queries per second for 800 million ChatGPT users by minimizing primary database load, optimizing queries with PgBouncer, and preventing cascading failures through systematic optimization rather than sharding. <a href="https://app.daily.dev/posts/0QBTpoQPy" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI is changing software development roles</strong>: AI coding tools are shifting developer roles from pure coders to orchestrators and validators, requiring new infrastructure for responsible AI-assisted development and emphasizing trade-off decisions and domain expertise over code output. <a href="https://app.daily.dev/feed-by-ids?id=2XGWtlWnL&amp;id=JV1i29nbu&amp;id=WYcCi2eVO&amp;id=Wvk48DEZ7&amp;id=dVvuHhcvo&amp;id=oZ8mOxinB&amp;id=qtVi65Njq&amp;id=G48nTExd2" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>NVIDIA claims massive performance gains for agentic AI</strong>: NVIDIA’s Blackwell Ultra platform (GB300 NVL72) reportedly achieves up to 50x higher throughput per megawatt and 35x lower cost per token for agentic AI workloads compared to Hopper, driven by hardware and continuous software optimizations. <a href="https://app.daily.dev/posts/XLkxKsrjj" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New architectural pattern for AI-driven software</strong>: A new ‘Deterministic Core, Agentic Shell’ architectural pattern is emerging, where LLM-based decision-making replaces traditional imperative control flow at the ‘shell’ layer, fundamentally shifting software logic branching. <a href="https://app.daily.dev/feed-by-ids?id=pbRQZ4xYf&amp;id=wVbWM6nJH&amp;id=xfda4p1KO" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Small language models for local deployment</strong>: Seven Small Language Models (SLMs) from 1B to 9B parameters are now available for local execution on consumer laptops, offering specialized performance for tasks like RAG, code generation, and edge computing. <a href="https://app.daily.dev/posts/Uggt1hDI5" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Agoda open-sources API agent for natural language queries</strong>: Agoda released API Agent, an open-source MCP server that connects to any REST or GraphQL API, introspects schemas, and generates queries from natural language input, using DuckDB for in-memory SQL post-processing. <a href="https://app.daily.dev/posts/VhIIJvhkg" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PostgreSQL diagnostic toolkit released</strong>: postgres_dba 7.0, a psql-based diagnostic toolkit for PostgreSQL, was released with 34 reports for bloat analysis, vacuum monitoring, index health, query performance, and more, requiring no extensions and compatible with versions 13-18. <a href="https://app.daily.dev/feed-by-ids?id=A4pYssSEb&amp;id=sS4pUElCH" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI-generated code is still mostly useless</strong>: Turns out Claude’s AI-generated C compiler can’t even compile ‘hello world’ on most Linux distros and makes SQLite queries run 158,000x slower than GCC. But hey, it can draw SVGs with a pen plotter, so that’s something. Clearly, we’re still a ways off from AI writing ‘essentially all code’ as some predicted. <a href="https://app.daily.dev/feed-by-ids?id=QCdIy73we&amp;id=ijXBeCZ92&amp;id=oHgSGtpkl&amp;id=qpa1K5xV7&amp;id=wQAJlLNnu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>The side-project SaaS dream is dead, blame AI</strong>: Apparently, the dream of building a profitable side-project SaaS is dead because AI-generated competitors flood every niche faster than you can ship. The author claims quality no longer matters, and the only winners are datacenter providers. So, instead of building better products, we’re just supposed to give up because AI can churn out mediocre alternatives? Sounds like a convenient excuse for not innovating. <a href="https://app.daily.dev/posts/dIaDDgpAV" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>KPMG partner uses AI to cheat on an AI exam</strong>: A KPMG partner was fined for using AI to cheat on an internal exam about AI. Because nothing says ‘expert in AI’ like needing AI to pass a basic test on the subject. This incident, along with other AI misuses at consulting firms, really instills confidence in their ‘AI fluency frameworks,’ doesn’t it? <a href="https://app.daily.dev/posts/RAmInAYNd" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
