Most of the currently used AV software contains heuristic/ML components that are able to gather and analyze various indicators of compromise. Without something like that running, you're basically tied to manual review of systems running at the moment. Making malware for such scenarios is basically making a Base64-encoded script in the language of your choice and then exploiting something(either the user or some software) to get it to execute.

I know, because I've been writing small malware toys and it got blasted by both Bitdefender and ESET.