this is like every other message on the gemini pro model. why do i get charged for laziness?
this is an issue gemini has with the edit tool. hopefully this is fixed in the new gemini release. When It happens sometimes it is easier to apply a manual edit, or temporarily swithc to claude 3.7. I di swing between the two and recently enjoying o4 mini althoug much slower but more reliable in complex context.
Gemini struggles a lot with the edit tool for some reason
Happens a ton for me as well. edit_file tool and even read_file tool are failing from time to time. When I start seeing failure modes like this too often. I ask for a summary or audit of what I’m (it’s?) working on and then click new chat including the chat history for a little while and then delete that attachment. I recently changed to taking that detailed summary or Audit digest and putting it in a file with my agent instructions to include on every chat until that part is complete. When I (it) completes a feature That when I audit or summaries detailed in a new direction. The main thing is when it starts getting super common… new chat is the way but try to keep some context from the history.
When we get maybe the 5/6/25 2.5 pro that might make a big difference since tool calling got better for google canvas with that one.
edit: I also use "you should try again with more specific instructions to the edit_file tool. I’ve seen it retry multiple times and use that language back to me about specificity when it does end up succeeding and it’s worked a lot more often when I say it. Also it has reported small changes being an issue and also used “It seems the model is stuck. Let’s try applying the change again, perhaps with slightly different phrasing.” before succeeding. So tldr, encourage the model for specificity and diversity until success lol…