Grammar and spelling mistakes are learnable problems to overcome. There are quite a few machine learning services you can use to aid bots, and there are also opportunities for a human to step in and instruct the AI when it can't properly determine intent (thereby teaching the bot for next time).

I don't know about your banking app, but there is no 3-step solution to finding out how much I've spent at Starbucks in the last month. I've tried something similar, and it's actually quite a few steps to try do an advanced search. This is a very limited example though. There are just a ton of financial questions out there that people have who probably don't ask because of the difficulty. Having banks provide this kind of service would be really exciting.

Certainly there is a place for a standard UI, but it will always be limited, otherwise the interface will be entirely too complex.

The biggest hurdle (for me at lease) doing banking this way is security. I would want my iPhone to ask for a fingerprint every time, just like using my app.

Though I'm not an expert, I've been reading everything I can on bots in the past year, and have created a few basic bots. I'll try to answer point-by-point.

1) In the not-so-distant future, there will be no "setting up" a bot. Here is a good video on how Skype envisions the interaction: https://www.youtube.com/watch?v=To7extbJRIA (there is a good example for coffee a few minutes in). The "agent" acts to facilitate the bot discovery process. And for the record, I cringe every time I go to use a pizza website or app (they are always horrible experiences imo).

2) Look at my answer in #1, particularly the coffee example in the video. The "Hey Pizza Bot!" is probably just for show in the screenshot.

3) I'd agree here, it might very well be better to actual state what the usual was. For me anyway I'd probably forget, but everyone is different I suppose.

4) This is important to understand. I haven't used Microsoft's Cognitive Services (https://www.microsoft.com/cognitive-services), however I can tell you that yes, to derive intent, you can certainly tell if someone meant "no" in many different ways. And the more that the bot gets used, the more it will learn. It won't be perfect obviously, but it's pretty good now, and will only get better. The systems I've used also give simple interfaces to steer uncommon phrases to certain intents.

5) Again, for a more complete example, take a look at that video. When the platforms really mature, I would fully expect that we will see a combination of text and simple UI interaction. (example: http://www.theverge.com/2016/3/30/11331168/klm-facebook-messenger-boarding-pass-chat-integration)

The big picture is hard to sometimes see, but I think it becomes much more clear when we consider the "agent" and "bot" relationship. The "agent" (Cortana, Google, Siri, Facebook M, Alexa, etc.) is breadth, and the bot is depth. It's incredibly interesting when you factor in that they will eventually talk and understand each other. I'll make a bold statement and say that this bot technology (though not sure if I like "bot" really) will surpass even the web in importance.

Hard to predict. We could all be writing applications for a messaging services in 2 years.

This is brilliant marketing from Burger King. Unfortunately McD's is handling it pretty poorly.

In my opinion, it's in everyone interest to develop and use standards. I don't think we are losing expressiveness.

It would be interesting to compare with trends of design in other media.

Being an "Expert" doesn't necessarily mean you know everything about a subject. I don't think there is anything wrong with referring to yourself as a web expert.

A lot of good questions here. Certainly, the web is becoming "just another app" on the internet, but how big will it end up being? It's fun to speculate.

It's not so far fetched. Why would you get your news from Facebook? It's a odd implementation though, definitely. It's so distinct from the main use of the app, without any integration that I could see. Maybe the pieces will come together later?

I think it's common to equate the "web" with the internet. The web is just a layer on top of the internet, but it is one of many (though the most popular at the moment).

Siri needs the internet to work.

That's why I quantified websites with "as they used to be". I guess my argument is, when we reach for the crutch (the hamburger menu, complicated dropdown menus, etc.) then we are avoiding a larger architecture issue.

Compare uk.gov and apple.com. You can't tell me that apple.com is more complicated. Uk.gov just did a much better job of designing/architecting for mobile devices. Apple.com stuck with the status quo. I think that's largely because of marketing, but that's just a guess.