What is the current landscape on AI agents knowledge

Recently used "free" rates codex to give me a quick fastapi project sample. It gave me deprecated (a)app.on_event("startup). What are your experiences on current AI agent code outputs. Doesn't have to be codex or claude or co-pilot. Whichever one you use just want to gauge your experiences on outputs as of 2026 Q1/Q2. Does the latest model always use the latest code documentations? questions: 1. I didn't specify which version of fastapi to use for output, do you type that everytime for your workflow? does it work if you specify like "use only the latest version" 2.