The biggest win for me has been cross-stack context switching. I maintain services in TypeScript, Python, and some Go, and the cost of switching between them used to be brutal - remembering idioms, library APIs, error handling patterns. Now I describe what I need and get idiomatic code in whichever language I'm in. That alone probably saves me 30-40 minutes on a typical day.

Where it consistently fails: anything involving the interaction between systems. If a bug spans a queue producer and its consumer, or the fix requires understanding how a frontend state change propagates through API calls to a cache invalidation - the model gives you a confident answer that addresses one layer and quietly ignores the rest. You end up debugging its fix instead of the original issue.

My stack: Claude Code (Opus) for investigation and bug triage in a ~60k LOC codebase, Cursor for greenfield work. Dropped autocomplete entirely after a month - it interrupted my thinking more than it helped.