My codebase grew to 40000 SLOC, not counting tests, and i increasingly had the feeling that codex wasn't keeping up with the complexity. When i sent it to rummage through the codebase for clean ups, code reviews, refactors or generally assessing the health of the codebase, i was under the impression that it didn't reach every codepath. That did improve with the switch from 5.4 to 5.5, but still not perfect. A little while ago i started instructing codex to "use x subagents". Depending on how wide-reaching the search would be, i used an appropiate number of subagents. Lately, i default to 6 subagents for these kind of tasks.
In my experience, that greatly helped codex to get the bigger picture and uncover lingering issues it wouldn't have caught without the use of subagents. It finds more relevant issues, does a better job of planning and, in general, gets a better holistic understanding of the codebase.
So if you get the feeling codex gets lost in the woods: don't despair, call subagents to your rescue!
"Disclaimer": i am running 5.5 with xhigh all the time, every time.