If anyone is thinking about switching to GitHub Copilot due to Cursor restricting slow-pool users from accessing Claude 3.7 Sonet and forcing them to use older models, here’s my experience.
I got a GitHub Copilot Premium account and tried using it to continue building my website, which I had previously been developing with Cursor + Claude 3.7. I also attempted to implement new features in my tools app, but I found Copilot with Claude 3.7 to be frustratingly slow and not very intelligent.
Copilot frequently asked about the codebase and files on every prompt, making the development process inefficient. Editing files took much longer than expected, and I encountered a lot of errors while making changes. Overall, Copilot did not provide the seamless experience I was hoping for.
After struggling with it for a while, I finally switched back to Cursor with Claude 3.5 Sonet, which has been a much better experience for me.
But at this point, we are completely at the mercy of Cursor, hoping they don’t impose further limitations.
I’m always testing alternatives, especially now that Cursor is unstable, but I generally come to the conclusion that Cursor is superior for the moment. I have, in addition to Cursor, VsCode Insiders with Copilot, Cline and Roo Cline, which have the advantage that, being extensions of vscode, they can be installed and tested in parallel on the same instance.
I think you should also give Cline and Roo Cline a try. If you already pay for GitHub Copilot, you can configure the use of Sonnet and the available models in an “unlimited” way there, configuring in API Provider: VS Code LM API, and being able to access claude-3.7, claude-3.7-thought, o1, o3-mini, etc., not only in Copilot but in Cine / Roo Cline. I never tried Codeium / WS (incredibly the word is censored in this forum ), but they have very good reviews too.
Anyway, I still come to the conclusion that Cursor is ahead, but it’s good to have viable alternatives, especially now, so as not to interrupt the workflow, but I hope that the problems will be solved soon and the final stable version 0.46 will be released on Linux. Trying so many things when you already have a workflow with a software wastes quite a few hours too.
Edit: Also, I think that Cursor should urgently review the way they send releases, I’ve been using Cursor for several months and it happened too many times, and the fact of automatic updates or rollbacks is something that should not exist. In that aspect I never saw something like that in any of the alternatives. Hopefully after this failed release this will be reviewed and fixed permanently.
WS is quite good,almost at the same level, just subscribed to them after cursor treated me(pro with slow pool) the same way they treated the trial abusers.
I tried Codeium / WS due to some hanging issues in Cursor (later found it is largely due to VS Code specific issues), but I couldn’t live without Cursor even for a day. Cursor is now like your committed life companion. You can’t leave.
Cursor is not buggy or slow for me. It costs maybe $30 extra to be a part of the fast pool. I literally replaced $300,000 of salaries with a $500 per year Cursor subscriptions. Insane.
Im pretty happy with the actual cursor on osx. Everything Works fine.
The only problem is sometimes after updates things working unexpected different and it’s hard to figure out it’s a bug or a feature. So i read the documentation after every update. A better communication about the changes might be nice.
Two salaries (200, 100) replaced, backend and frontend. The answer to your question is long, it wasn’t easy, but I know enough now to do it well. I’m editing a massive codebase and I have less bugs than ever, and I’m adding a feature a day. I’m not experiencing bloat because I know what to look for and how to prompt it. Things that would take my devs a month are done in 1-2 days now myself. I’ve also improved security and stability. The software is better than ever thanks to Sonnet 3.5 and now 3.7 taking over. If you know what you are doing, and you work with Cursor for HOURS non-stop (And figure out how to get through all of the frustrating parts and unsolvable loops that you eventually solve), you learn that it’s like having a senior dev at your fingertips. And believe it or not, even when I had my devs, I was telling them where/how to fix things based on what AI was telling me and they’re like “wow how did you know?” lol
Yesterday I tried Roo-Code with Anthropics api.key ! Why ? For one week I was trying to implement an authentication system ! It took me 8 days trying to fix it to work with Cursor AI and always failing and every time making it even worst! I tried with Roo-Code yesterday and took me JUST ONE SHOT! This is dramatic difference!! My only problem is that the api key of the anthropic has rate limits and that very sad!
I’ve also experimented with different tools and I think in these fairly early days of agentic based development it’s normal to keep switching between different tools. I use mainly Copilot, Cursor and RooCode at the moment. So far I do think Cursor fits my style a lot better and I’ve made some really cool things with it. But similarly I’ve made some great apps with Copilot and it costs a lot less but lacks many of the features other tools have and the rate limiting on there can be a bit annoying.
What I really like about Cursor is its ability to load docs or use the web for searching. The docs feature in particular has been a game changer being able to target lesser known frameworks or SDKs improving overall accuracy. I wish Copilot would add that feature soon.
), but they have very good reviews too.