<h2>Must know</h2>
<ul>
<li><strong>TypeScript overtakes JavaScript and Python on GitHub</strong>: For the first time, TypeScript is the most used programming language on GitHub, surpassing both Python and JavaScript. This 67% year-over-year growth is attributed to developers favoring typed languages for more reliable AI-assisted coding in production, according to GitHub’s latest report. This shift highlights a growing industry preference for robust, type-safe codebases, especially as AI tools become more integrated into development workflows.</li>
<li><strong>AWS launches massive AI supercomputer and multimodal embeddings</strong>: AWS has activated Project Rainier, a colossal AI supercomputer with nearly 500,000 Trainium2 chips, already training Anthropic’s Claude models. Concurrently, Amazon Nova introduces multimodal embeddings for improved Retrieval-Augmented Generation (RAG) accuracy and Web Grounding for real-time, citation-based retrieval. These developments, reported by AWS and The Register, position AWS as a major player in the AI infrastructure race, offering advanced tools for AI model development and deployment.</li>
<li><strong>New open-source tools challenge Terraform and Jupyter</strong>: Platform Engineering Labs launched Formae, an open-source infrastructure-as-code tool that automatically codifies existing infrastructure, aiming to simplify cloud management beyond Terraform. Meanwhile, Deepnote, a collaborative data science notebook platform, has gone open source, positioning itself as a successor to Jupyter by addressing its collaboration and AI integration limitations. These tools, highlighted by InfoQ, signal a growing trend towards more automated, open, and AI-friendly development and infrastructure management.</li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>AI coding assistants expand across web frameworks</strong>: Syncfusion launched AI Coding Assistants for Blazor and React, with plans to support Angular, .NET MAUI, Vue, and JavaScript. These assistants, leveraging Model Context Protocol (MCP) servers, integrate into VS Code and Code Studio to generate context-aware code and component configurations, promising to accelerate development.</li>
<li><strong>Bun 1.3 simplifies JavaScript development</strong>: Bun 1.3 introduces frontend tooling integration, native SQL support for Postgres, MySQL, and SQLite, a new Redis client, and single-file executable bundling, aiming to streamline JavaScript app development and deployment.</li>
<li><strong>Elixir 1.19 boosts type system and performance</strong>: Elixir 1.19 enhances type inference for anonymous functions and protocols, and significantly improves compilation speed for large projects by up to four times, ensuring compatibility with Erlang OTP 28.1 and achieving OpenChain certification.</li>
<li><strong>Google Cloud enhances GKE for AI workloads</strong>: Google Cloud launched managed DRANET in GKE for topology-aware scheduling of network interfaces with GPUs, reporting up to 59.6% bandwidth improvements. Additionally, open-source Ray on Kubernetes received updates for flexible scheduling, dynamic resource allocation, and in-place pod resizing, optimizing GPU/TPU usage for distributed AI/ML workloads.</li>
<li><strong>Shadcn UI evolves with CLI 3.0 and new registry</strong>: Shadcn UI, a code distribution platform, released CLI 3.0 for faster tooling and introduced a new Registry Directory for component discovery. This platform continues to offer full customization and ownership of UI components, with ongoing updates to form primitives.</li>
<li><strong>Go 1.25 introduces experimental ‘Green Tea’ garbage collector</strong>: Go 1.25 features the experimental ‘Green Tea’ garbage collector, which reduces GC overhead by 10-40% by processing entire memory pages and leveraging vector instructions, with plans to make it the default in Go 1.26.</li>
<li><strong>Ubuntu optimizes AI models for diverse hardware</strong>: Canonical released silicon-optimized generative AI models (DeepSeek R1, Qwen 2.5 VL) as Snap packages for Ubuntu 24.04 LTS. These packages automatically detect underlying hardware like Intel, Ampere, and NVIDIA GPUs to deploy the most efficient inference stack, simplifying AI deployment on Ubuntu.</li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI-powered debugging: the magic bullet for bugs?</strong>: LogicStar AI claims its platform can autonomously investigate, reproduce, fix, and validate software bugs with 90% precision accuracy, reducing bug lifecycles by 95%. Apparently, it uses deep static and dynamic analysis to build ‘rich application context’ for ‘cheaper LLM queries.’ So, instead of hiring competent engineers, we can just throw more AI at the problem and expect our code to magically fix itself. What could possibly go wrong?</li>
<li><strong>AI-powered artifact repositories: because DevOps wasn’t complicated enough</strong>: JFrog Fly introduces the ‘world’s first agentic artifact repository’ for AI-native development teams. It promises to automatically collect and manage packages, containers, and models with natural language commands, all while maintaining ‘control and trust.’ Because what every developer needs is another layer of AI abstraction managing their releases, especially when it’s ‘zero-configuration’ and built on a platform that ‘delivers enterprise DevOps practices without requiring dedicated DevOps teams.’ Sounds like a recipe for delightful debugging.</li>
<li><strong>Australian police want to translate emojis with AI</strong>: The Australian Federal Police are developing an AI tool to interpret emojis and Gen Z/Alpha slang in encrypted communications to combat online criminal networks. Because apparently, the biggest threat to national security is a 😂 or a 🍆. While they’re at it, maybe they can teach the AI to understand sarcasm, unlike some of the ‘experts’ pushing this stuff. I’m sure this won’t lead to any hilarious misinterpretations or privacy concerns whatsoever.</li>
</ul>
