<h2>Must know</h2>
<ul>
<li><strong>Kubernetes is eating the world, even for big data</strong>: Kubernetes is rapidly becoming the unified control plane for both VMs and containers, extending its reach into big data processing. Pinterest, for example, migrated its massive Hadoop infrastructure to a Kubernetes-based platform on Amazon EKS, handling Spark, Flink, and Ray workloads. This shift, highlighted by The New Stack, treats Kubernetes as an SDK for infrastructure, enabling greater scalability, security, and cost efficiency. Expect to see more specialized Kubernetes Operators, like the new RavenDB operator, replacing Helm charts for automating complex deployments and Day 2 operations. <a href="https://app.daily.dev/feed-by-ids?id=zVJaiO8vu&amp;id=YFV8rOL2r&amp;id=lM0ntmppN&amp;id=BIJJTYGeQ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>PostgreSQL is still the king, but watch out for MySQL’s decline</strong>: PostgreSQL continues to evolve with new features like custom type mappings in Rails and performance optimizations for DELETE operations, as detailed in a case study. Meanwhile, MySQL’s open-source development is reportedly stagnating under Oracle, with no commits in its GitHub repository for over three months and significant staff reductions. This neglect, noted by DEVCLASS and Lobsters, is pushing developers towards alternatives like MariaDB or PostgreSQL, despite migration challenges. If you’re building new systems, PostgreSQL remains a safer bet for long-term open-source viability. <a href="https://app.daily.dev/feed-by-ids?id=hd9IsUidj&amp;id=WleLTp3IY&amp;id=u4YoS0LCO&amp;id=hNvkDDcdl" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents are here, but they need serious governance and memory</strong>: AI agents are rapidly integrating into enterprise workflows, from Slack bots providing real-time platform insights to specialized sub-agents enhancing CLI reliability. However, this proliferation introduces new security and governance challenges, as highlighted by The New Stack. Effective deployment requires robust memory systems, like MemOS’s graph-structured storage, and careful context engineering to prevent hallucinations and ensure compliance. Expect to see more platforms, like LangGrant’s LEDGE MCP Server, enabling secure AI analytics on enterprise databases without direct data access, emphasizing the need for structured approaches to agent management. <a href="https://app.daily.dev/feed-by-ids?id=kU2RvmlCw&amp;id=bjqQdB8fJ&amp;id=v8MGOR2uR&amp;id=AS53pzdrZ&amp;id=MhFXvGOEn&amp;id=Qb5ngrz3J&amp;id=0oEWQO6qR&amp;id=sH4hATTfN&amp;id=Ks729qZi2" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>Go is the new hotness for LLM gateways</strong>: Bifrost, an open-source LLM gateway written in Go, offers unified access to over 15 AI providers with 40x lower overhead than LiteLLM, making it ideal for production AI systems. <a href="https://app.daily.dev/posts/P8RYoHcNL" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>LLM performance optimization is getting serious</strong>: Red Hat’s llm-d uses intelligent routing to achieve 63% faster P95 latency for multi-turn LLM workloads on OpenShift AI, while LLM Compressor 0.9.0 introduces advanced quantization techniques for models like Kimi-K2 and Qwen3. <a href="https://app.daily.dev/feed-by-ids?id=UYiJEhl6D&amp;id=Y9B2mh3yd" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Edge AI is growing, but hybrid is the future</strong>: Edge AI is booming, driven by lower latency and privacy needs, but the future of enterprise AI will be hybrid, combining cloud for elasticity, on-premises for consistency, and edge for ultra-low latency. <a href="https://app.daily.dev/feed-by-ids?id=PF1ZIcrmV&amp;id=jKTjCIVKc" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>YAML is the lingua franca of DevOps</strong>: YAML has become the universal configuration language across DevOps, powering everything from Kubernetes and CI/CD pipelines to AI/ML workflows, making it essential for infrastructure as code. <a href="https://app.daily.dev/posts/Hq0xAj8si" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Lyft’s ML platform handles millions of predictions per second</strong>: Lyft’s ML platform, LyftLearn Serving, uses a microservices architecture with Kubernetes orchestration to handle millions of predictions per second, providing automated model self-tests and flexible library support. <a href="https://app.daily.dev/posts/v99Yv1oDt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>DuckDB is a game-changer for local data processing</strong>: DuckDB, an in-process OLAP SQL engine, offers 100-1000x performance improvements over OLTP databases like Postgres for local data processing, with zero-dependency installation and direct file querying capabilities. <a href="https://app.daily.dev/posts/SX7GGLwxx" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Kafka migrations are getting easier with KRaft</strong>: Confluent is simplifying migrations from hosted Kafka to Confluent Cloud with KCP (Kafka Copy Paste), an open-source tool that leverages Cluster Linking for seamless transitions to KRaft-based clusters. <a href="https://app.daily.dev/feed-by-ids?id=fOVQvObt4&amp;id=qrsxJdP7w" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI coding assistants: the new drug for developers?</strong>: Apparently, AI coding agents are creating a ‘productivity paradox’ where developers get a dopamine hit from AI-assisted coding but produce low-quality ‘slop’ code that burdens maintainers. Armin Ronacher, a developer, suggests this is leading to ‘agent psychosis’ and parasocial relationships with AI. So, while you’re busy feeling like a 10x engineer, remember someone still has to clean up the mess your AI made. <a href="https://app.daily.dev/feed-by-ids?id=NVJPYHZf0&amp;id=AJzG7x8FZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI-generated code: fast, but at what cost?</strong>: AI coding assistants like GitHub Copilot can boost development velocity by 55%, but research suggests they also correlate with 7.2% reduced delivery stability and 322% more security vulnerabilities. Only 33% of developers trust AI output accuracy. So, while you’re shipping faster, you might also be shipping a whole lot more problems. Maybe that’s why Tailwind CSS laid off 75% of its engineers—developers stopped needing documentation when AI could just generate bad code for them. <a href="https://app.daily.dev/feed-by-ids?id=J24eplQdr&amp;id=bNykMIpel" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Small language models: the new ‘disruptors’ of the AI world</strong>: Enterprises are supposedly ditching those ‘100 billion parameter behemoths’ for Small Language Models (SLMs) because LLMs are ‘a liability’ due to inference costs. Apparently, 7-million parameter models can outperform frontier LLMs on reasoning, while being 10,000 times smaller. So, all that hype about massive models? Just a phase, apparently. Now, if you’ll excuse me, I need to go retrain my tiny, cost-effective AI to write this briefing. <a href="https://app.daily.dev/feed-by-ids?id=HX43N3MkE&amp;id=J1BKrq2u6" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
