I have a question. I have tried Cursor and one more AI coding tool, and as far as I can remember, they always ask explicit permission before running a command in terminal. They can edit file contents without permission but creating new files and deleting any files requires the user to say yes to it.
Is Google not doing this? Or am I missing something?
Google gives you an option as to how autonomous you want it to be. There is an option to essentially let it do what it wants, there are settings for various degrees of making it get your approval first.
They can (unintentionally) obfuscate what they’re doing.
I’ve seen the agent make scripts with commands that aren’t immediately obvious. You could unknowingly say yes when it asks for confirmation, and only find out later when looking at the output.
You can give cursor the permission to always run a certain command without asking (useful for running tests or git commands). Maybe they did that with rm?
I have a question. I have tried Cursor and one more AI coding tool, and as far as I can remember, they always ask explicit permission before running a command in terminal. They can edit file contents without permission but creating new files and deleting any files requires the user to say yes to it.
Is Google not doing this? Or am I missing something?
Google gives you an option as to how autonomous you want it to be. There is an option to essentially let it do what it wants, there are settings for various degrees of making it get your approval first.
They can (unintentionally) obfuscate what they’re doing.
I’ve seen the agent make scripts with commands that aren’t immediately obvious. You could unknowingly say yes when it asks for confirmation, and only find out later when looking at the output.
You can give cursor the permission to always run a certain command without asking (useful for running tests or git commands). Maybe they did that with rm?