The current batch of models, specifically Claude Sonnet and Opus 4, are the first I've used that have actually been more helpful than annoying on the large mixed-language codebases I work in. I suspect that dividing line differs greatly between developers and applications.