Haven't we seen this future evolve for decades? FORTRAN used to be the tool of choice for any work that involved crunching numbers. Then VisiCalc came along in the late 70s / early 80s building on good ideas that originated in the 60s and took (aside from macros) the programming work out of number crunching for most laypeople. Those who still needed more under-the-hood capability got tools like MATLab.
Software loaders / installers started as byte loaders and moved along to what we have today which is often specified using config style files that don't involve a lot of programming, but stand on the shoulders of general-purpose installers that programmers wrote.
The trend is to take routine things, generalize them, and create tools to provide those generalized capabilities to people who then don't have to write code. So the trend is toward codeless software.
However, there's always a leading edge to all of this; some next new thing that needs to be created, that often stands on top of all of the generalized automations that we no longer have to write, leaving us time to think about the new good ideas. We can pull together tremendous amounts of functionality as we build out containers, but we still need to commit some code. IFTTT is a great example of automations being built without writing software, but someone wrote IFTTT, exposing its generalized tool to the masses in a way that non-programmers could automate routine tasks.
Yes, there's a future for codeless software, and hopefully we're writing it. But there's also a future for code pushing the boundaries.
From what we can see, the future Siri still phones home at the soonest opportunity to get the request parsed. The request is almost certainly then outsourced to a coding and art team from sometime in the early '90s (by sling-shotting a courier ship around a star to deliver the task details, obviously), and the results are incorporated into the install files for the ship's computer so it is available immediately upon the original request.
I wanted to say something about how no matter how good computers are at understanding what humans mean, humans are still consistently bad at meaning they want. And something something, leaky abstraction something (... I didn't fully format that argument in my head). But it's already mentionned in the linked discussion.
Although, I don't see any reason why sufficiently advanced graphical programming wouldn't be able to match written code. It would require graphical elements that are more abstract and precise though, which would need to be understood. Technically what "codeless" is actually implying is "learnless" or "little learning required", which is where the actual roadblock is.
I once had a wise boss who said that he once thought that computers should do what we mean rather than what we say. After some reflection, he decided that this would not be much better. What we need is a machine that can "do what is appropriate." I suspect that requests to such a device would be called "prayer."