Cursor0.50 is a major update, one step closer to AI native IDE, and its ability to handle complex projects is greatly improved

Cursor released the 0.50 update today, which is a fairly large version change and worth paying attention to. I have further sorted out the information and compiled a clearer, easier to read and more complete web page: cursor05.bookai.top

In short, Cursor 0.50 has adjusted the billing model, improved Max Mode capabilities, and introduced a very special workflow (Background Agent). I think this is an important node in Cursor's evolution towards an "AI native IDE". The following is a detailed explanation of the seven key aspects of this update.

1. Update Overview

2. Billing model: simpler and more intuitive

Version 0.5 completely restructures the billing system, abandons the cumbersome settings, and switches to a more intuitive request billing model:

1. Normal mode: Pay per request

• Each AI interaction is considered a request
• Regardless of the size of the context, each interaction consumes a fixed number of requests
• Suitable for common interactions and simple tasks in daily development

For example: Using Claude 3.5 Sonnet for one interaction consumes 1 request

2. Max Mode: Billing by token

• Similar to the billing method of API calls, charging is based on actual token usage
• Contains all tokens for input, output, file reading, and tool calls
• Suitable for handling large code bases or complex tasks

The Pro version includes 500 free requests per month

For most everyday coding tasks, it is more economical to use Normal mode.

Switch to Max Mode for critical tasks such as refactoring and important bug fixes

3. Max Mode is fully open

Previously limited to certain models, Max Mode is now fully open to all top AI models, making high-context processing capabilities universal:

Project scale references for different context sizes:

• 10K tokens
  : Can handle small utility libraries
• 60K tokens
  : Suitable for medium-sized feature sets
• 200K tokens
  : Can cover the complete web framework level code
• 1M tokens
  : Core code capable of handling large frameworks

To use it, turn on the Max Mode switch from the model picker.

It is recommended to use Max Mode when analyzing a project for the first time to establish a global understanding

With the @folders command, you can achieve true "global code understanding"

4. Background Agent: Taking AI collaboration one step further

This is the most groundbreaking feature of Cursor 0.5, which allows AI to truly work continuously in the "background" and achieve task processing without human intervention:

Core Competencies:

• Asynchronously execute complex tasks in a remote containerized environment
• Support multiple concurrent agents to process different tasks at the same time
• After the task is completed, PR can be automatically submitted or the user can be notified
• No manual supervision or confirmation is required at each step

Applicable scenarios:

• Only supported in Max Mode, works with all Max Mode compatible models

• GitHub has been integrated, and PRs can be created directly on remote repositories

• The more precise the task description, the more the execution results will meet expectations.

5. Context and editing experience optimization

Cursor 0.5 also brings a number of improvements to improve programming efficiency in daily use:

1. Codebase context improvements

• @folders command enhancements
  : You can put the entire code base into context
• Contextual Status Icons
  : Visually display whether the file is included in the context
• Smarter file selection
  : Automatically include as many files as possible that match the restrictions

2. Inline Edit interface update

• New and simpler interface design
• New Full file edit option (⌘⇧⏎) to edit the entire file without a dialog
• Added Send to agent option (⌘L) to send the current task to the agent for processing

3. Improved editing speed of long files

• Introducing the search & replace tool to directly locate the exact location that needs to be modified
• Precise editing without reading the entire file
• In the Postgres code base test, editing speed increased by nearly twice

4. Workspace Management

• Support multi-root workspaces, handle multiple code bases at the same time
• All code bases will be indexed, and AI can search for references globally
• Especially suitable for multi-warehouse collaboration scenarios such as front-end and back-end, main project and dependent library

5. Chat function enhancement

• Supports exporting conversations to Markdown format for easy sharing and archiving
• Support for session duplication, making it easy to explore new scenarios while preserving existing conversations
• Agent now uses native terminals, providing a smoother operating experience

6. Best Practices

1. Create dedicated chats for different types of tasks to avoid context confusion

2. Use Max Mode + @folders at the beginning of the project to establish a global understanding

3. Small encoding tasks continue to use Normal mode to save request quota

4. For time-consuming or repetitive tasks, consider using Background Agent first

5. Use multi-root workspaces to manage multiple related projects

6. Frequently used code snippets and ideas are exported and archived in time to form a personal knowledge base

Recommended configurations for different scenarios:

VII. Conclusion

I personally feel that Cursor 0.50 will allow you to do better with large projects through Background Agent, greater context handling capabilities and simplified usage, although this will also bring increased costs.

With the addition of new functions, the complexity of some uses seems to have increased, but the boundaries of our project capabilities have also been improved. So just learn... just keep learning and you'll be fine.