[fogx]

> especially if those inputs are extremely vague (how on earth should we interpret slow?!)

isn't this exactly the (theoretical) strength of a chatbot - asking follow-up questions to remove uncertainty?