AI-assisted development moves into the editor, not the browser
AI-assisted development is the practice of embedding artificial intelligence directly into code editors so that code generation, refactoring, and project context management happen inside the same environment where developers type, debug, and run their software, instead of in disconnected browser tabs or external tools.
The new default for serious developers is clear: your code editor is no longer a text box with plugins; it is a command center for AI-assisted development. Visual Studio Code now ships on-device multilingual dictation and side chats in its 1.132 release, reducing the constant tab-hopping to separate chatbots or transcription tools and keeping developers in one place while they think and type. Zed, built in Rust with a GPU-accelerated UI, advertises itself as “an absolute speed demon” while folding strong support for AI agents directly into its core. And outside the editor, some developers are wiring Claude Code, Codex, and local LLMs into a shared Obsidian vault so every tool reads and writes the same project memory. The battle isn’t about whether you use AI; it’s about where that AI lives and who owns the context.
VS Code’s AI integration: powerful, but heavy and extension-led
VS Code is leaning hard into AI, but it is doing so on top of an extension-first, Electron-heavy foundation. The 1.132 release adds multilingual dictation and side chats directly in the editor, along with element-level feedback in the integrated browser and Markdown diffs in its hybrid Markdown editor. That means you can talk to your editor, keep an AI chat fixed to the sidebar, and let it comment on what the internal browser is rendering, all without leaving VS Code. For many teams, this is exactly why VS Code remains “one of the most popular editors for both beginners and experienced developers”.
The price is bloat. Out of the box, VS Code is fine, but as soon as you install the AI extensions that make it competitive, resource usage spikes: with extensions, it can consume 3 to 4GB of RAM. Because it rides on Electron and Chromium for rendering, graphics latency creeps in as more plugins pile up. That tension defines VS Code’s AI story: the editor that became famous for being light is now the heaviest place in many developers’ stacks, even while it tries to be the most convenient hub for AI-assisted development.
Zed: AI agents without the Electron tax
Zed takes the opposite stance: build a fast, minimal core, then treat AI agents as first-class citizens without leaning on Electron. Written in Rust and using its own GPU-accelerated GPUI framework, it renders UI directly on the graphics card and delivers 120 FPS scrolling. In internal benchmarks, Zed opens large, multi-thousand-file projects in roughly 0.25 to 0.6 seconds, compared with VS Code’s 1.3 to 3.8 seconds. Since its idle memory footprint is lower, developers can keep browsers, Docker, local servers, and AI agents running on machines with modest RAM without feeling squeezed.
The Parallel Agents update is the real inflection point. Released in April 2026, it adds a Threads Sidebar so you can run and monitor multiple agent sessions in one window. Zed’s Agent Client Protocol (ACP) gives external agents access to project context, multi-file editing, and change review. Claude, Codex, Gemini CLI, OpenCode, Copilot, Cursor, and more are already available through the ACP Registry. One quotable summary from the benchmarks is: “opening a large, multi-thousand-file project takes Zed roughly 0.25 to 0.6 seconds, compared to VS Code’s 1.3 to 3.8 seconds”. Zed includes many features that would require separate extensions in VS Code, and its extension ecosystem is still growing. If VS Code is the comfortable default, Zed is the opinionated bet that AI code completion and agents should not come with a 4GB RAM bill.
Shared knowledge bases: Obsidian as memory for multiple AI tools
While editors fight over who owns the AI experience, some developers are quietly shifting the power center to their notes. One developer wired Claude Code, Antigravity, Codex, and a local LLM into an Obsidian vault and turned that vault into a shared AI memory. They created a meta folder with five markdown files describing machine specs, vault organization, active projects, connected tools, and preferred agent behavior, then pointed each tool at those files before any work begins. They even set up pointer files like CLAUDE.md, AGENTS.md, and lm-studio-prompt.md so every agent knows where to read onboarding instructions.
The effect is that “every fresh session everywhere used to mean re-explaining who I am and what I'm doing, and now that's one file that tells the tool everything”. Claude Code can decide something about a landing page, Codex can understand that decision later, and a local Qwen 3.5 instance in LM Studio can summarize it again, all because they read and write the same Obsidian files via a local filesystem MCP. The vault outlasts any single editor or AI service: projects, notes, and configuration live together in one knowledge base that can onboard new agents automatically, and future tools like Antigravity projects can be pulled in without restructuring.
Cloud vs local AI: control, privacy, and the new baseline
Behind all these workflows is a quieter shift: local and on-device AI are no longer fringe experiments; they are becoming part of the baseline. On the local side, that Obsidian-powered workflow uses Qwen 3.5 running in LM Studio to read the same files as cloud tools and summarize project decisions through a local filesystem MCP. That means sensitive context can stay on disk while still feeding into AI-assisted development, and developers keep more control over where their project history lives. On the editor side, Zed’s low idle memory footprint makes running local agents alongside everything else realistic even on lower-spec machines.
Cloud tools are not going away. Zed’s ACP Registry lists Claude, Codex, Gemini CLI, OpenCode, Copilot, Cursor, and more. VS Code’s multilingual dictation and side chats show that cloud-backed features are now standard expectations, not luxury add-ons. The direction of travel is obvious: AI-assisted code completion and generation are table stakes, and editors that ignore them will be ignored in turn. The real decision for developers is strategic: do you centralize your workflow in a heavyweight, extension-driven editor like VS Code, move to a performance-first environment like Zed, or treat your personal knowledge base as the true hub and let AI tools orbit around it? For now, VS Code is still the safest bet for stability, but the smartest developers are already decoupling their memory from their editor and experimenting with lighter, more controllable stacks.
