start Claude Code in either VS Code or Cursor Terminal, then run /IDE in Claude Code. It tells Claude Code that it’s running in IDE… One bug though,- the shift+Enter for new line isn’t working on Linux machine. I still haven’t found a way to get it working… so if you do,- please share 
I used Claude Code since yesterday afternoon. Initially very frustrating,- but once you get it integrated properly and learn the few quirks,- works awesome! I’m on the the Max plan since I got loads of work for next few weeks… but even coding whole day and using Opus (it was a single agent running, since work couldn’t be split to sub-agents) I never hit limit (and it resets every 5 hours)
4 Likes
This feature (vector indexes) is amazing or not? I need your opinion.
I popped $15 on openrouter yesterday to use the free 1000 dailies. Still experimenting with all the free models.
I am running through vs code/Cline.
I didn’t see how to use openrouter with Cursor. Is this possible? GPT says it is but I think he’s tripping ballz.
Once the cursor fixes the API key, yes, it will definitely work. That setup gives you 5,000 daily questions on GLM 4.5, which opens just like Sonnet does—one per day, my friend. As soon as they fix it, it’s going to be amazing
I am still wondering if you are a troll or serious…
Your posts are hilariously naive. Keep it going! 
Obviously everyone who disagrees with you is just a troll. Maybe we know something you don’t. I laid of 10 devs this month because auto can do a better job.
sounds very interesting! please keep us posted about how you stick with auto and how this was a good decision! This might be hard to believe now, but I am honestly interested because we already had so many news about companies laying off devs only to heavily regret it or even go bankrupt. I honestly believe in a time where you will not need “devs” anymore but rather “product owners” or something similar but I honestly doubt that one person with auto can replace 10 devs. PLEASE prove me wrong in a month or three.
I use Cursor and Zed with openrouter, how ever , Cursor works more efficient for me. This efficiency lead actually to me burning more on openrouter than on cursor. Both using Claude Opus 4.1
Yes, but even today, if you notice, it still doesn’t work in the latest version.
I have a guess that with 5,000 questions per day you can do a lot with that, especially using glm 4.5
I have a guess that with 5,000 questions per day you can do a lot with that, especially using glm 4.5
But I can’t use it because the API is disabled.
My comment may be from a few days ago, but productivity has been welll over 10 times what it used to be for about six months now. 1 person is not more productive because they use auto. They are more productive because they are using auto on 50 different computers and electricity is cheap. Anyone who only has one instance of cursor open at a time is doing it wrong. Sadly we just don’t have enough work, for 11 people to be 10 times as productive so we laid off the other ten people. Gpu’s are expensive but devs are even more so.
can you tell me more about your setup?
- about the 50 different computers: are you using agents or something like RDP to actually run 50 instances of cursor on remote machines?
- are these 50 agents/cursor instances working on the same codebase at the same time?
how would you characterize the architecture of your code base, in case it is a single code base?
I would be very happy to learn how to further boost my productivity.
It is one code base at a time, I have different roles all with fixed prompts for each role. And a few agents whose only job is to write documentation fast enough to keep all the other agents busy. So the prompts are fixed and super simple, but the number of agents doing a task changes throughout the project.
Also fun fact english isn’t very dense in terms of information per syllable, so if you translate to different languages, you can get more data processed for the same price.
Thanks for sharing.
about the language: Agree, I also fail to see why no machine language is being used for reasoning. Makes no sense at all to have reasoning in natural language.
This is done because the content models are trained on is primarily natural language. Technically it would work with any language.
Of course it is trained on natural language but this would not stop it from using machine language, as it is done for its embeddings, inside the model, or has been shown to be working already for model-to-model communication.
1 Like
The first version of DeepSeek R1 wrote code in its mind. I think I’ve seen this a couple of times from Gemini or another model in Cursor.