Most older parsers kind of do this already, as they only look for the first few letters in a word. Zork used to only look for the first 6 characters, and the old Scott Adams games looked for only 3 or 4. You could type OPE CHE to open chest. A modern author could do something similar by putting in abbreviated synonoms.
In Inform at least there’s also a compile-time option to change this, although it usually isn’t necessary.
ADRIFT’s autocomplete is bad enough that it’s worse than playing with it off.
“s” is completed to “shove”, “i” is completed to “in”, “l” is completed to “lay”, when those should be left alone.
“b” is completed to “bind”, when “b” is a valid command.
A command involving a basket was completed to one basked (which I haven’t seen), when I was holding a different basket.
I don’t know what game you were playing rm, but it sounds like someone was playing around with the autofill priorities.
I’ve just checked, and in a normal game “s” is completed to “south”, “i” to “inventory”, “l” to “look”, and “b” to “blow”.
Autocomplete as a standard across all the various parser-game engines would be bada**! I think it’s a little bit archaic that it’s not commonly used yet. It makes sense to me that it would only autocomplete things already seen (in the current room prioritized), items in your inventory, people you’ve met, and verbs you’ve already used.