Zubnet AI學習Wiki › Cursor
公司

Cursor

一個 AI 原生的程式編輯器,作為 VS Code 的 fork 建構,圍繞「程式編輯應該是和 AI 模型的對話」這一想法設計。Cursor 把 LLM 深度整合到編輯體驗:行內程式生成、多檔案編輯、程式碼庫感知的上下文、自然語言指令。它是 AI 編碼助手更大浪潮的一部分,和 GitHub Copilot、Windsurf、Claude Code 一起。

為什麼重要

Cursor 代表一個賭注:AI 將從根本上改變程式是怎麼寫的 — 不只是自動補全建議,而是 AI 作為理解你整個程式碼庫的協作者。它的快速採用(一年內數百萬使用者)和開發者回報的生產力提升,讓它成為 AI 改變知識工作今天最具體的例子之一。

Deep Dive

Cursor's technical approach centers on context: the editor indexes your codebase, identifies relevant files and symbols, and includes them in prompts to the LLM. This codebase-awareness is what separates it from generic chat interfaces — when you ask "refactor the auth module to use JWT," Cursor knows which files contain the auth module, what the current implementation looks like, and what other files depend on it.

The Editor vs. Plugin Debate

Cursor's decision to fork VS Code (rather than building a plugin) was controversial but strategic. A fork lets them deeply integrate AI into the editing experience — custom keybindings, inline diff views, AI-aware file navigation — in ways that a plugin's limited API surface doesn't allow. The trade-off is maintaining compatibility with VS Code's extension ecosystem while diverging on core features. Competitors like GitHub Copilot and Claude Code take different approaches: Copilot as a VS Code extension, Claude Code as a terminal-based tool.

相關概念

In The News

Cursor's Warp Decode Claims 1.8x GPU Speedup—But Where's the Proof?
Apr 07, 2026
Cursor Doubles Down on AI Agents After $3B+ Funding Round
Apr 03, 2026
Cursor Fights Back Against AI Labs With New Agent Interface
Apr 02, 2026
Cursor Claims 99% Speed Boost for AI Code Search, But Details Are Thin
Mar 24, 2026
Cursor Backtracks After Using Chinese Kimi Model Without Disclosure
Mar 22, 2026
See all 6 articles about Cursor →
← 所有術語
← Curriculum 學習ing Data Augmentation →