Cursor ditches VS Code, but not everyone is happy...

Cursor — think of it like Microsoft Word, but for writing code — just released version 3.0, and it's a complete rebuild (Source 3). Until now, Cursor was based on VS Code, Microsoft's free code editor that most developers use. Cursor simply added AI features on top of VS Code's foundation, like adding a calculator app to your phone's existing software.

Now Cursor has thrown that foundation away and built its own editor from scratch using Rust (a programming language known for speed) and TypeScript (Source 3). It's like if someone took apart a Honda Civic engine and rebuilt the entire car around a Ferrari engine instead.

The big change: Cursor 3.0 isn't just autocompleting your code anymore. It now runs multiple AI agents — think of them as virtual assistants that can each work on different parts of your code simultaneously — across "multiple repos, machines, and cloud environments" (Source 3). Before, Cursor's AI suggested one line at a time. Now it can theoretically rewrite your entire project while you sleep, with different AI agents handling the database, the user interface, and the API at the same time.

Cursor also released Composer 2, their own AI model built specifically for writing code (Source 3). Instead of relying entirely on OpenAI's GPT or Anthropic's Claude, Cursor now has its own "brain" trained on coding tasks. This sparked controversy — developers are debating whether Cursor's model is actually better or if this is just a marketing play (Source 3).

The problem: ditching VS Code means ditching VS Code extensions — the plugins developers have spent years customizing. If you've been using a specific linter, theme, or debugging tool, it probably doesn't work in Cursor 3.0 anymore. Some developers who tried the update reported broken workflows and are sticking with Cursor 2.x or switching back to plain VS Code (Source 2, Source 3).

The real story here is Cursor betting its entire product on a future where developers don't write code — they manage AI agents that write code. If you're still thinking of AI coding tools as "fancy autocomplete," you're already behind. Cursor 3.0 is designed for a workflow where you tell an AI "build me a login system with OAuth and rate limiting" and it generates 500 lines of code across 12 files while you make coffee. That's not science fiction — that capability exists today (Source 3).

But here's the uncomfortable truth: most developers don't actually want that yet. They want AI to speed up boring tasks (writing boilerplate, fixing bugs) while they keep control of architecture decisions. Cursor's pivot to multi-agent orchestration assumes developers are ready to hand over the keys. The backlash suggests many aren't (Source 3). For comparison, when GitHub Copilot launched, it stayed safely inside VS Code and let developers ignore it if they wanted. Cursor is forcing a choice: adopt our vision of AI-first development or leave.

Why this matters even if you don't code: the tools developers use determine what software gets built and how fast. If Cursor's bet pays off, apps you use daily could be built 10x faster by small teams using AI agents. If it fails, we'll learn that humans aren't ready to let AI make architectural decisions — even when the AI is technically capable. Either way, this is the first major editor to burn its bridges with the old model and go all-in on agents. Watch what happens next.