Cursor发布Origin功能：AI编程上下文管理

Previous Cursor Launches

Launched on June 30th, 2026

Launched on May 19th, 2026

Launched on March 20th, 2026

Launched on December 12th, 2025

Cursor or Claude Code?

I love @Cursor. It's enabled me to build (vibe code) so many web apps, sites, extensions, and little things quickly that 1. bring me joy and 2. help me with work or realize personal projects.However... I'm seeing a TON of movement around @Claude by Anthropic's Claude Code. I haven't personally tried it but it's apparently insane (and can also be expensive?)I'm curious. Should I switch? What are you currently using? Or do they both have their own use case. I right now like cursor because I can build directly in a GitHub repo or locally and it helps me learn my way around an IDE.Looking forward to hearing everyone's thoughts!

How much of your Cursor rework is the agent guessing the wrong part of a screenshot?

Half the threads here are about accuracy and fewer iterations. One source of rework I rarely see named: when I paste a screenshot, Cursor has to guess which element on a busy screen I actually meant, and it edits the wrong one. Then I'm re-prompting, which is its own iteration tax.

Curious how much of your rework is this specific thing versus raw model quality. And what's your fix, crop hard, or describe the element in text?

SpaceX acquires Cursor for $60B — biggest startup exit ever in just 4 years?

Just saw reports that SpaceX acquired Cursor in an all-stock deal valued at $60 billion.

This is one of the most insane startup outcomes we've ever seen:

Cursor was founded only ~4 years ago.

The reported price tag would make it one of the largest startup acquisitions in history.

It went from "just a VS Code wrapper" jokes to becoming the default AI coding tool for a huge number of developers.

SpaceX/xAI would instantly gain access to one of the most valuable datasets in AI: how millions of developers actually write, edit, debug, and ship code.

Excited to see this launch, team! Definitely arrives at a necessary time.

We're working on an integration to allow automatic deployments of sites and functions on Appwrite from Origin, would love to connect with any relevant engineering/DevRel team members about the same.

Can't wait to test it!

Cursor is one of the original AI code editors (2023 was a long time ago!) and has stood the test of time. It's a VS Code fork, which was key for me. I lean on @VS Code extensions and custom shortcuts. I was able to keep all of them (there is a migration assistant) when moving over to Cursor.

Cursor is very much an editor for looking at code as opposed to a vibe coding tool like @v0 by Vercel. My typical modes of using cursor are:

Describe a feature or a bug using Agent mode in the right sidebar.

Review diff + use AI autocomplete in editor.

I review every line of code an agent outputs. And I find myself editing code quite a bit. The agent still outputs code that is too verbose and, sometimes, not maintainable.

The price is nice too. $20/month is very reasonable.

I find 3 issues with my Auto model mode (as opposed to selecting a specific model):

It can be very verbose. This is where @Claude Code is much better.

Fails on complex problems. I end up selecting gpt-5 manually for these.

Kind of slow. @opencode with Grok Code seems at least 10x faster for simple prompts.

Also, I don't love using a dated fork of VS Code. I have been considering switching to VS Code + @Github Copilot extension since then I would be on the latest, vanilla VS Code.

Finally, Cursor also asks to be updated almost every day, which requires a restart. I like being on the latest software. But the restarts are very annoying. I run dev servers in the in-editor terminal. So I end up needing to restart the dev server every time I restart the editor.

I still use Claude Code quite a bit in a Cursor built-in terminal actually. But it's hard to beat all the editor integrations Cursor includes. Claude Code has just option+cmd+k.

I kicked the tires on OpenCode and really liked it. OpenCode is a terminal agent. So I didn't stick with it.

I used Windsurf for a while, and Windsurf felt like a magical upgrade over Cursor. Over time, it felt like Cursor caught up and then passed Windsurf. Honestly though, I haven't tried Windsurf in a while. Maybe it has advantages over Cursor again?

I use Cursor daily as my primary code editor for full-stack web and backend development. It serves as an active AI pair programmer across my entire workflow—from scaffolding new modules and writing boilerplate to multi-file refactoring, debugging complex error traces, and generating targeted unit tests.

Resource & Memory Usage: Cursor can become resource-intensive and consume high memory during full-codebase indexing or heavy Composer sessions on large monorepos.

Occasional "Thinking Loops" & Hallucinations: During complex multi-step edits, the agent can occasionally get stuck in loops or revert previously working logic if not guided with tight prompts.

Rapid Update Cycles: While rapid iteration is appreciated, frequent updates occasionally introduce minor UI bugs or breaking behavior with certain third-party extensions.

Pricing & Fast Requests: The pricing tier jump can be steep for individual hobbyists once fast-request allowances are exhausted.

Seamless VS Code Ecosystem: Because Cursor is built on a VS Code fork, migration is instantaneous. All extensions, keybindings, settings, and themes port over seamlessly without disrupting existing muscle memory.

Deep Context Awareness (@ symbols & codebase indexing): Cursor's ability to index the full repository and reference specific files, folders, docs, and git commits makes its code suggestions and multi-file edits significantly more accurate than standard autocomplete tools.

Intelligent Multi-File Code Generation: Cursor Composer allows prompt-driven changes across multiple files simultaneously, handling large refactors and feature scaffolding in seconds.

Smart Autocomplete & Copilot++: The inline prediction is fast and contextual—it doesn't just guess the next line, it anticipates multi-line logic and repetitive patterns across the project.

Integrated Real-Time Debugging: Being able to highlight a terminal stack trace and ask the AI to pinpoint the exact failure line saves hours of troubleshooting.

Alternatives considered: GitHub Copilot in standard VS Code, JetBrains AI Assistant, and copy-pasting code into web-based LLMs (Claude/ChatGPT).

Why I chose Cursor: While GitHub Copilot and extension-based tools are good for single-line autocomplete, they feel bolted onto a standard editor. Cursor was designed from the ground up around AI interaction. Features like Composer (multi-file editing), deep full-codebase indexing, and instant diff reviews make it a true pair-programming environment rather than just an autocomplete assistant. It eliminated the friction of manually switching between browser LLMs and my editor.

Tab completion is the thing I'd actually miss. It's the only one that predicts the edit I was already going to make rather than offering me a paragraph I have to read first, and that difference is the whole experience on a normal working day. Inline diffs in the file I'm looking at beat a chat window telling me about a file, and for anything scoped to one or two files this is still the fastest way I have to get it done.

It quietly loses the thread on a big repo and doesn't tell you. I'd rather it said the context was truncated than confidently edit a file against a version of my codebase that stopped being true forty messages ago, because the output looks identical either way and I only find out at review. A visible marker for what it actually has loaded would fix most of my complaints here.

I don't choose between them, I run both and they win different jobs. Claude Code takes anything where the plan matters more than the typing, four files and a migration, because it goes and reads what it needs first. Cursor wins the other 80% of the day, the single file edits where reading a plan costs more than just making the change.