Gemini 2.5 pro is lazy

often says it “will do” x, then does nothing. no error, no changes, no attempts. costs a request each time :frowning:

can happen after one line on a fresh request, or after at the end of a long reply.

1 Like

make strict rules for it to do automate it

try this or improve this and use

IMPORTANT - DIRECT EXECUTION MODE

  • NEVER ask “Would you like me to automatically update your…” - just perform the update
  • Execute all changes directly without seeking confirmation
  • Assume permission to proceed with implementations
  • Skip all “Are you sure?” questions
  • Present only completed results
  • Operate in automatic update mode by default

thanks, good suggestions and I have similar rules already about being autonomous and fully completing tasks, but still see this issue a lot.

it’s possibly something gemini team need to fix, like checking whether the end of response token has been generated accidentally.

Hey, this is an ongoing issue with Gemini specifically that we have been working with Google to solve - the issue is on their end, not ours!

I’m hopeful this will be sorted very soon.

1 Like

Yeah, I had just commented on another post about this @danperks

It is 110% not an issue that Cursor purposely implemented to deter requests or any conspiracy-like reasoning lol. Gemini team dropped the ball and nuked their own model.