A Complete Collection of the Best Tips for Using Cursor

Guides
1

Recently, I complained about the new pricing model and reverted to the old one ($20/month for 500 requests). Things have been quite smooth. Two workdays and 54 fast requests, everything was running well :heart_eyes:

In fact, if you want to use Cursor to continuously run Vibe Coding projects with just $20/month, it might be a bit tricky. You’ll need to understand and control your codebase to use the AI optimally :smiling_face_with_tear:. Also, creating new projects repeatedly will make the AI generate a lot of files and tool calls :face_exhaling:.

However, for day-to-day programming tasks, the $20 plan works really well. Regardless of who you are, knowing how to use the tool correctly is essential for maximum efficiency.

Here’s how I usually work with Cursor:

  1. Create a few well-defined instructions and include them in the rules, but don’t overdo it-especially avoid behaviors that lead to frequent manual corrections (this helps the AI reduce the need to “learn” everything :hot_face:, lowers tool calls, trims unnecessary context, and reduces retries).
  2. Keep each conversation focused on closely related tasks, and ask the AI to perform several actions at once :sparkles: (this is effective when those functions share similar context-significantly reducing the number of AI calls).
  3. Avoid overly long conversations. Irrelevant context builds up. Once a set of related tasks is done, start a new thread. This significantly improves quality because the context becomes more focused and also reduces the prompt size sent to the AI :magic_wand:
  4. Continuously refine your prompts and rules based on how the AI performs-so each iteration is better than the last.

There are a few metrics to help assess whether you’re using AI efficiently:

  1. Input tokens < 50K (ideally under 20K-roughly 10 files, each with 200 lines of code).
  2. Output tokens < 30K (since output often includes lengthy explanations, guide the AI not to over-explain. Don’t tell it to skip explanations entirely. This way you can generate 5–7 files of 200 lines each).

Unfortunately, with the old pricing model, Cursor no longer shows me the token usage stats (I hope they bring that back soon :pink_heart:).

The project I’m currently working on is similar to ClickUp (an all-in-one work platform), with a massive and complex codebase. The way I use Cursor has been quite effective, and I hope it helps someone out there.

If you have any tips to improve quality, reduce costs, or save time, please share them! :smiling_face_with_horns:

3 Likes