<h2>Must know</h2>
<ul>
<li><strong>AI coding assistants get smarter with MCP servers</strong>: Vaadin and Docker Desktop are rolling out Model Context Protocol (MCP) servers, allowing AI coding assistants like GitHub Copilot and Claude to access documentation and tool integrations directly from your editor. This means more accurate, context-aware code suggestions and easier setup for AI tools, reducing hallucinations and manual configuration. Docker’s approach isolates each MCP server in a container, provisioning only when needed to save resources. This is a big step towards making AI assistants genuinely useful for developers, as reported by Vaadin and Docker’s official announcements.</li>
<li><strong>Major cloud outages highlight resilience failures</strong>: AWS suffered a significant US-EAST-1 outage, impacting services for hours due to a DynamoDB DNS race condition. This cascaded, affecting critical services like pharmacies in Norway and even Anthropic’s Claude AI. Industry experts, as noted in various technical analyses, are stressing that current cloud redundancy strategies aren’t cutting it, and organizations need to focus on comprehensive dependency mapping and real-time observability. It seems even the big players can’t escape the inevitable, and your contracts probably won’t save you from the business damage.</li>
<li><strong>AI infrastructure spending goes parabolic, but where’s the ROI?</strong>: Tech giants like Meta, Google, Microsoft, and Oracle are pouring billions into AI datacenters, with Meta alone issuing $30 billion in bonds for expansion. AWS reported 20% revenue growth, citing AI demand. Yet, despite $776 billion invested from 2023-2025, most companies aren’t seeing significant AI service revenue, with Microsoft leading at a mere $13 billion ARR, largely from discounted OpenAI compute. Analysts are questioning the sustainability of these investments, especially as enterprises struggle to get strong ROI from their own AI projects, according to financial media.</li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>TypeScript now GitHub’s most used language</strong>: GitHub’s 2025 Octoverse data shows TypeScript has surpassed Python and JavaScript as the most used language, with 67% year-over-year growth, attributed to developer preference for typed languages in AI-assisted coding.</li>
<li><strong>AI agents get new security and scheduling features</strong>: Amazon Connect now allows individual agent schedule merging, while Aembit released an IAM solution for ‘Agentic AI’ with cryptographically verified identities and ephemeral credentials, and AWS Marketplace supports new AI agent pricing models.</li>
<li><strong>New tools for building AI agents and workflows</strong>: DeepAgents CLI offers persistent memory agents for coding and research, while Pipelex provides an open-source language for creating repeatable AI workflows with typed steps and multi-provider support.</li>
<li><strong>Real-time data pipelines are crucial for production AI</strong>: Outdated data pipelines, not model quality, are causing AI failures, with Apache Kafka and Write-Ahead Log systems (like Netflix’s and DoorDash’s) becoming key for sub-millisecond latency and continuous feature engineering.</li>
<li><strong>Kubernetes and AI infrastructure management advances</strong>: Cover Whale improved microservice deployment with Kratix, Google Cloud’s Managed Lustre offers high-performance storage for AI inference, and KAITO/KubeFleet manage AI workloads across clusters, with Google releasing a GKE Gemini CLI extension for natural language cluster management.</li>
<li><strong>Frontend development sees new tools and performance boosts</strong>: VoidZero launched Vite+ for unified JavaScript tooling, shadcn/ui released CLI 3.0, and Bun 1.3 introduced frontend tooling and improved Node.js compatibility, while Next.js added Cache Components for better client navigation performance.</li>
<li><strong><a href="http://Mozilla.ai" target="_blank" rel="noopener nofollow">Mozilla.ai</a> takes over llamafile for local LLM execution</strong>: <a href="http://Mozilla.ai" target="_blank" rel="noopener nofollow">Mozilla.ai</a> is modernizing the llamafile project, enabling single-file distribution and local execution of LLMs with a focus on privacy and community-driven development, though cloud LLMs still generally outperform local models in speed and quality.</li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI is transforming risk management from reactive to proactive, apparently</strong>: Google Cloud is pushing the narrative that AI is magically transforming risk management from reactive to proactive, enabling predictive analytics and automated controls. Sure, because throwing AI at legacy infrastructure and talent shortages will just magically fix everything. They even endorse integrating AI risk management with NIST and ISO frameworks, as if a fancy framework will stop the next Tata Motors-level breach. It’s almost like they’re trying to sell you something.</li>
<li><strong>AI-powered bug bounty programs are a mixed bag of ‘slop’</strong>: HackerOne reports a 210% increase in valid AI vulnerability finds, which sounds great until you realize the volume of ‘AI slop’—low-quality, AI-generated submissions—is straining triage capacity. So, AI helps find bugs, but also creates a mountain of garbage for humans to sort through. It’s almost like autonomous agents are still unreliable for context-sensitive flaws, and we’re just trading one problem for another.</li>
<li><strong>AI is making developers ‘write docs, not code,’ says Google Cloud</strong>: Keith Ballinger from Google Cloud is now arguing that developers should ‘stop writing code, start writing docs,’ claiming agentic AI tools work best with detailed context. He predicts increasing abstraction will eliminate routine code review as developer input moves towards natural language. So, instead of writing actual code, we’re all just going to be ‘specification engineers’ feeding the AI beast. Sounds like a great way to lose touch with how things actually work.</li>
</ul>
