Because Amazon didn't code it.
It would be relatively simple for Amazon to make a Routine action that we type in the text command we want it to execute and it parses.
well, kind of. Amazon's API is oddly written such that one variation of that concept, a company takes text, runs it through text-to-speech and feeds the audio into Amazon. But somewhere, deep inside the code stack, is an engine that takes text and does stuff with it.
I suspect the simplest reason they didn't do that in Routines is security. A Routine can't lock a door. It won't run a scene on my SmartThings that includes locking the door. Because Amazon wouldn't touch door locks, probably to avoid liability. If they ran plain-text, then they'd be touching a door when I write "ask ADT to lock the front door."
There's probably a cascade of concerns, like that one, with opening the door to any command you can say. Including covering what they can't think of. The "we don't know what hole there could be, but since the possibilities become vast, let's allow that" argument.