It’s a bit of an odd one really Wza. (to be clear, I’m writing with the perspective of someone who’s worked in voice for a decade or so now)
Long post hidden for those who aren’t interested; click if you’re deathly bored 
Summary
Alexa, Google Assistant, Siri and such like aren’t really up to much on their own. We know this because not many people use them as the main way of using their phones, and as you’ve pointed out yourself, the AI is a bit rubbish. I still get my commute coming up on my phone even if I’ve put an all day event in my calendar saying I’m somewhere else.
The devices (Google Home, Echo, Apple Home Pod) are nowt more than interfaces into Alexa, GA, Siri etc. There’s some decent technology going on in terms of directional listening, noice cancellation and suchlike, but not much more than that. All they’re doing otherwise is giving you a way of accessing a not hugely functional bit of AI.
What IS good about them (and in particular the Google version) is the quality of ASR (voice recognition) going on. Having worked with ASR, they’ve pushed the boundries of what is possible a long way, very quickly, and they’re all giving developers everywhere the ability to tap into that - something that you used to have to go to Nuance for and pay hundreds of thousands in licencing and server costs to set up for yourself.
However, most developers don’t understand how to build voice user interfaces that are useful or usable, because everything we’ve done with technology until recently has been visual and precise (we give users a limited toolset to do what they want - for example, to save something you can click the save button, use the file menu or hit ctrl + s). We’ve all experienced it with telephone systems, when (for example) it takes 15 steps to book a cinema ticket because it only prompts you for one bit of info at a time (and half the time it mishears you).
Voice is not a precise interface and developers/designers without experience in developing specifically for voice are notoriously bad at building systems for it. However, it can do great things.
Take that cinema example again. I could build an Alexa skill that can book you a cinema ticket in a few seconds. Obviously you’d need to set it up with some details initially - preferred cinema, seating arrangement, membership/payment details - but once I’ve got that it’s really simple
“Hey Alexa, ask Cineworld to book me a ticket for Toy Story 4 on Thursday after 7pm.”
“Okay, I can get your usual seat for a 7:15 screening in Leicester Square. Is that okay?”
“Yup”.
Away from home? Override the default cinema in that initial question.
“Hey Alexa, ask Cineworld to find me a screening of Planet of the Apes tomorrow evening near Colwyn Bay”
“The nearest Cineworld is in Llandudno and will be showing The Wrath of the Planet of the Apes at 8:30 tomorrow. Would you like me to book that for you?”
“Yes please”.
Note that even though the wording of the question was unclear (it didn’t specify the cinema, OR the correct name of the film, the stuff behind the scenes worked out the intent of the user)
Now, we obviously have mobile apps and websites that can do this kind of booking, and it gives you more control if you do it yourself - maybe you don’t know what you want to watch or at what time of day, in which case voice is a horrible UI for imparting that amount of information. Leaving aside the point someone made above about accessibility though, using voice for moderately complex tasks if the app/skill is designed properly can be much quicker and often simpler cognitively for a user. Moreover, it doesn’t tie you to any device… If you’re in the kitchen baking and you remember something you meant to do, then just speaking is much easier both cognitively and timewise than washing your hands, finding your phone, unlocking, finding the right app, completing your task and going back to you baking again.
TL:DR It’s not necessary for most people and it’s not that useful at the moment because many “skills” don’t understand use cases or VUI properly, plus it will likely never take over entirely (as already said, it’s dreadful for browsing) but there are a lot of potential use cases where it will shine in the future.
