How does context buildup work for the code generating machines generally ? Do the programs just use human notes + current code directly ? Are there some specif

How does context buildup work for the code generating machines generally ? Do the programs just use human notes + current code directly ? Are there some specific ranking steps that need to be done ?