Hey - how many of you are taking the cursor advice and starting a ‘new chat’ when prompted when using Gemini 2.5 Pro.
I’m curious because the message to start a new chat comes up around the same time as it would if I was using, say, Claude but the context window is much larger with Gemini so I wonder if it is really necessary to start a new chat?
I have heard of people saying they let Gemini run for dev sessions of over an hour. That scares me. What do you do?
This is hard to specify without the context of a project.
Depending how many files are needed for a task, how complex the project is, how detailed the requirements are, how much code needs to be written in a framework or language…
I have seen some filling up a full context before they even start a single request. While others work with a single context for quite long but not an hour 
What it comes down is experience. You need experience with the model, with your project and prompting.
A while back I tried Gemini 2.5 Pro and found it not working well for me as a model due to its behavior that is so different from Claude Sonnet models.
Then a few days ago I tried again with a more relaxed approach and it did a great job.
With time you get comfortable knowing when a new chat needs to be started way before Cursor suggests a new chat because of early model hallucinations. Then you also see when you do not need to start a new chat to complete a task.