I see LLMs as a force multiplier. It's not going to write entire projects for me, but it'll assist with "how do i do x with y" kind of problems. At the end of the day I still understand the codebase and know where its faults lie.