<h2>Must know</h2>
<ul>
<li><strong>AI coding tools are still a mess, but they’re here to stay</strong>: AI coding assistants like Claude Code and Cursor are generating code faster than humans, but often with significant quality issues and security vulnerabilities. While some claim AI can build a browser from scratch in a week, developers are finding these tools struggle with greenfield projects, DevOps, and test code, often producing verbose and generic solutions. The real challenge isn’t just code generation, but the increased review burden and the need for human expertise to guide the AI and clean up its mistakes, as highlighted by Hacker News and other developer experiences. This means developers need to focus more on architecture, planning, and critical evaluation rather than just typing code.</li>
<li><strong>Public APIs are closing up shop thanks to AI</strong>: Remember when public APIs were all about open data and innovation? Well, those days are fading fast. Companies like Twitter, Facebook, Google Maps, Strava, Spotify, OpenAI, and Marvel are increasingly restricting or shutting down their public APIs. Why? Because AI models and agents are scraping vast amounts of data cheaply, making it hard for these platforms to monetize their offerings or protect their intellectual property. This shift, reported by Nordic APIs, means less interoperability and more proprietary AI features, fundamentally changing the landscape for developers and third-party services.</li>
<li><strong>NVIDIA and AMD push AI hardware and software boundaries</strong>: NVIDIA is making waves with its TensorRT for RTX, bringing adaptive inference to consumer GPUs for automatic optimization of AI models, as detailed by NVIDIA Developer. Meanwhile, AMD is expanding its ROCm 7.2 platform with support for new GPUs and beta visualization tools, and vLLM-SR is demonstrating a Mixture-of-Models architecture on AMD MI300X and MI355X GPUs for intelligent LLM routing. These developments signal a fierce competition in AI hardware and software, aiming to make AI inference faster and more efficient across various platforms.</li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>TypeScript backend frameworks get AI-ready</strong>: A roundup of TypeScript backend frameworks shows that tools like Encore, with its built-in MCP server, can generate complete applications with AI in just 6 minutes, significantly faster than traditional methods, by focusing on AI integration and real-time feedback loops.</li>
<li><strong>GitHub CLI integrates Copilot for seamless AI assistance</strong>: GitHub CLI now directly supports <code>gh copilot</code>, replacing the deprecated extension and streamlining the use of the agentic Copilot CLI for developers, making AI-assisted coding more accessible from the command line.</li>
<li><strong>React Native for Windows gets an update, but questions remain</strong>: Microsoft released React Native Windows 0.81 with Hermes debugger support, but it lags behind mobile versions and drops C# support, raising questions about its long-term strategy compared to MAUI.</li>
<li><strong>New BigQuery gen AI functions for better data analysis</strong>: BigQuery now offers generally available AI.GENERATE and AI.GENERATE_TABLE functions, along with new preview features AI.EMBED and AI.SIMILARITY, enabling multimodal data processing and semantic analysis directly within SQL.</li>
<li><strong>Prisma 7 goes Rust-free for performance gains</strong>: Prisma ORM 7.0 has replaced its Rust-based query engine with a TypeScript implementation, resulting in smaller bundle sizes, faster query execution, and improved developer experience.</li>
<li><strong>Superplane launches as open-source DevOps control plane</strong>: Superplane, an open-source control plane for event-driven DevOps workflows, integrates with various tools for CI/CD, monitoring, and incident management, offering features like approval gates and progressive rollouts.</li>
<li><strong>OpenSearch Kubernetes Operator 3.0 alpha boosts stability</strong>: The OpenSearch Kubernetes Operator 3.0 Alpha introduces significant stability improvements, including quorum-safe rolling restarts, multi-namespace support, and TLS certificate hot reloading, addressing critical production issues.</li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI coding assistants: the ultimate productivity hack (for creating tech debt)</strong>: So, AI coding assistants are supposedly making developers 10x faster, but apparently, that speed comes with a side of 7.2% reduced delivery stability, 322% more security vulnerabilities, and a plummeting refactoring rate. It’s almost like blindly generating code without understanding the underlying architecture or having proper guardrails just creates more problems than it solves. Who knew that treating AI like a magic junior developer would lead to a mountain of technical debt? What a shocker!</li>
<li><strong>AI agents: the new frontier of enterprise security risks (because we totally learned from the last one)</strong>: With AI agents now outnumbering human identities by 80 to 1, and operating with broad, delegated permissions, enterprises are facing a ‘new category of insider threat.’ Apparently, these autonomous marvels are exposing ‘previously unseen weaknesses’ in cloud infrastructure, like inadequate network segmentation and insufficient identity-based security. It’s almost as if giving AI agents unchecked power without proper governance or accountability is a recipe for disaster. But hey, at least we’re innovating, right?</li>
<li><strong>The great AI documentation paradox: can AI even read its own instructions?</strong>: AI assistants are supposed to make our lives easier, but apparently, they can’t even read documentation properly. According to Arcjet, tools like GitHub Copilot execute JavaScript and strip CSS-hidden content, while ChatGPT reads raw HTML without running scripts. So, to make sure AI can actually understand our docs, we need to ‘server-render critical content’ and ‘minimize reliance on client-side JavaScript.’ It’s like we’re going back to the 90s just so our cutting-edge AI can comprehend what we’re trying to tell it. Peak efficiency, folks.</li>
</ul>
