Hi everyone,
I’d like to hear about your real-world experiences using Gemini 3 Pro (High), especially around which tasks it handles well and which ones you prefer to do manually.
In particular:
-
What types of tasks or workflows do you think Gemini 3 Pro (High) excels at?
-
In which cases does it tend to struggle or introduce friction?
-
What practical tips or prompting strategies have you found useful to get more reliable results?
I’ll briefly share my experience so far to set the context.
In my case, when using Gemini as an agent for software development tasks:
-
I’ve noticed a relatively high error rate.
-
It often loses or partially forgets prior context.
-
It frequently rewrites code unnecessarily, even when the project already contains existing classes or abstractions that solve the same problem.
-
As a result, I usually need 3 to 5 iterations before getting a working solution that actually fits into the existing codebase.
Because of this, I’m still evaluating where it adds the most value versus where manual implementation is faster or safer.
I’d be very interested in learning how others are using Gemini 3 Pro (High) effectively, what boundaries you set for it, and any best practices you’ve developed.
Thanks in advance for sharing your insights.