Is it bad to start containers in the custom executor config stage?

I need to start a container and add its ID to the job_env so all other steps, especially cleanup, can access/interact with that instance/container without making API calls to get all running containers and then iterate until I find the one that matches some value. However, the docs hint, though not strictly, that this should be done in the prepare stage of custom executors. Is there any reason why can’t do it in config?