🤝 AI Coding Editor as Everything App?
Dear curious mind,
AI coding editors are no longer exclusive tools for writing code, they are becoming central hubs for managing files, integrating with external services, and orchestrating workflows through agentic AI capabilities. This issue explores how these editors are redefining productivity.
In this issue:
💡 Shared Insight
AI Code Editors Evolve Towards Your Next Operating System
📰 AI Update
Qwen3: Outstanding Open-Weights Models for Every Device from Alibaba
Phi-4-Reasoning: Compact and Powerful Reasoning Models from Microsoft
Zed: Agentic Workflows in Open-Source AI Code Editor
💡 Shared Insight
AI Code Editors Evolve Towards Your Next Operating System
Last week, I touched in the Media Recommendation section how AI Code Editors are evolving beyond their traditional roles and becoming powerful tools for more than just coders. Building on my talk at the PKM Summit on "Supercharge Markdown Notes with AI Code Editors," where I demonstrated using AI code editors like Cursor, Windsurf and Zed for advanced Markdown file management, it is clear these tools offer vast potential. They allow AI-powered semantic search, advanced file edits, and integration with MCP servers, extending capabilities far beyond native note-taking apps without needing extra plugins.
This week, I want to highlight my favourite editor, Zed, an open-source AI code editor which is not well known in comparison to Cursor and Windsurf. In contrast to these, the editor is not a fork of Visual Studio Code, but was coded from scratch in the programming language Rust, which is well known for its high performance and memory safety. Zed natively supports running models locally via Ollama and LM Studio on your own device, offering greater control and privacy. The latest updates to Zed are particularly exciting as they introduce agentic capabilities and simplify the integration of MCP servers. Especially the latter pushes the boundaries of what these editors can do, as any tool can be wrapped to be used by the AI.
This progression brings us closer to a fascinating vision articulated by Andrej Karpathy back in November 2023: the concept of an "LLM OS."
Using an AI coding editor equipped with MCP servers feels like a tangible step towards this "LLM OS." Think about your daily computer usage: You are either working with files or interacting with software tools and services, whether local or cloud-based. An LLM integrated deeply into your editor could orchestrate all of this. It could generate not just text and code, but also audio and video, with the file system as output medium. Integrating software tools and services as MCP servers gives the LLMs running in your coding editor new capabilities. The idea of an editor becoming a central hub for everything you do on your PC is incredibly powerful and something I'm eager to explore further. To be honest, this idea is not new, just have a look at the classical editor Emacs, which is well-known for its extensibility. The following joke is as old as the editor itself:
Emacs is an OS that lacks a decent text editor
The strategic importance of this space is underscored by recent industry moves. You might have come across the news which confirmed OpenAI’s acquisition of the AI coding editor Windsurf for a 3 billion dollars. This significant investment by a major AI player raises compelling questions:
Do they share this vision of an AI coding editor evolving into an "LLM OS"?
Or is their interest primarily in the user base and the valuable training data it generates?
What will be the fate of Windsurf? Will it continue as a dedicated application, or will its development shut down as OpenAI integrates its capabilities into ChatGPT?
Only time will tell. However, this scenario highlights the immense benefit of open-source solutions like Zed. An open-source tool has a much higher chance of survival and continued development, even in the event of an acquisition of its commercial counterparts. My fingers are crossed that the Zed development continues as it currently does and becomes the central tool for PC interaction I envision. Being open-source also empowers users to contribute directly to this vision by modifying and extending its codebase.
The journey of AI coding editors from specialized developer tools to potentially central operating interfaces is one of the most exciting developments in AI. It's a path worth watching and, for the adventurous, actively participating in.
📰 AI Update
Qwen3: Outstanding Open-Weights Models for Every Device from Alibaba [Qwen blog]
The Qwen team from Alibaba released Qwen3, a suite of open-weight large language models available in multiple sizes—from 0.6B to a 235B (Mixture-of-experts). The various sizes enable deployment on consumer laptops, desktops, and even embedded devices like phones. The performance of all model sizes is outstanding and competes with the currently best closed models. Furthermore, the models have improved agentic capabilities and a better support for MCP (Model Context Protocol). I love it! To test the models yourself, you can run them via Ollama or test them initially at the officially Qwen website.
Phi-4-Reasoning: Compact and Powerful Reasoning Models from Microsoft [Microsoft blog]
Microsoft released the 14-billion parameter Phi-4-reasoning model and the 3.84B Phi-4-mini-reasoning model as open weights. Their surprisingly strong performance on reasoning tasks, considering their relatively small sizes, is impressive. The accompanying technical report provides a detailed overview of the model development, from their supervised fine-tuning to the addition of reinforcement learning. Microsoft's stated that their intention is to enable local execution of these models on their Copilot+ PCs running Windows 11, showcasing a commitment to accessible, powerful AI on personal devices.
Zed: Agentic Workflows in Open-Source AI Code Editor [Zed blog]
Version 0.185.9 introduces officially the Agent Panel, which I already explored in the private beta. The integration of MCP servers as tools is now as easy and straightforward as in Cursor or Windsurf. However, in contrast to the competition, you can use the agentic capabilities with local models in a privacy-friendly setup.
Disclaimer: This newsletter is written with the aid of AI. I use AI as an assistant to generate and optimize the text. However, the amount of AI used varies depending on the topic and the content. I always curate and edit the text myself to ensure quality and accuracy. The opinions and views expressed in this newsletter are my own and do not necessarily reflect those of the sources or the AI models.