I wholeheartedly agree; the app-centric model of today's desktop environments is the antithesis of the Smalltalk vision of composability. In an environment where everything is a live object, programmers can send messages to those objects. Thus, in the Smalltalk environment you end up with something even more powerful than Unix pipes and redirection.
Now, Smalltalk provides the infrastructure for composability, but Smalltalk by itself doesn't provide us the full-fledged desktop environment and applications that users have come to expect; they would have to be implemented in Smalltalk. But where things get interesting in a Smalltalk-implemented desktop environment is that the live object environment is still there, leading to interesting possibilities. Imagine being able to control a word processing program through scripts written outside the word processor, without the word processing program itself having to implement something like Visual Basic for Applications? Imagine being able to control a spreadsheet with Smalltalk methods on cells instead of having to learn the spreadsheet's language or having to learn Visual Basic for Applications? The possibilities become even more intriguing when objects can interact with each other in ways that are far more general than Unix pipes. This is the desktop environment I dream of using, the unification of the ease of use of GUIs and the flexibility and programmability similar to the Unix shell.
Oddly enough, in the mid-1990's Apple implemented a less-ambitious (but still very ambitious for its time) version of this vision known as OpenDoc (https://en.wikipedia.org/wiki/OpenDoc). There are some wonderful videos at https://www.youtube.com/watch?v=oFJdjk2rq4E (a three minute summary) and at https://youtu.be/2FSFvEIpm5o (a 50-minute demo). OpenDoc was released in 1996, if I recall correctly, and there were some applications written with OpenDoc, most notably the CyberDog web browser (https://en.wikipedia.org/wiki/Cyberdog). However, once Apple bought NeXT in late 1996, Apple committed itself to building its next-generation operating system on NeXT, which had its own collection of object-oriented APIs. The cancellation of OpenDoc led to this famous spat during WWDC 1997 when a disgruntled developer questioned Steve Jobs on Apple's decision to cancel OpenDoc (https://www.youtube.com/watch?v=oeqPrUmVz-o).
Why was OpenDoc cancelled? One can argue that OpenDoc's cancellation was due to Apple needing to have a tight focus during its very vulnerable period in 1997. The proof is in the pudding: mid-1990's Apple before the return of Steve Jobs was an unfocused beacon that was able to produce many interesting technological demos (the Dylan programming language, SK8, OpenDoc), but there was no single coherent vision. The Taligent and Copland projects, which strove to replace the classic Mac OS with then-modern underpinnings, were disasters. OpenDoc was just one out of the many, sometimes competing, visions that Apple had during this time. Steve Jobs was able to turn around Apple's fortunes by having Apple focus on one vision for the Mac.
However, another way of looking at the cancellation of OpenDoc is that its success would have completely upended the software industry. Instead of software vendors selling apps, software vendors would sell components, which users can integrate to create custom workflows. While I believe that this would have led to a lot of flexibility and user-empowerment, OpenDoc would have also been seriously challenged by major software vendors like Microsoft and Adobe, whose empires were built on selling large, proprietary software packages. They were not going to give up their moats without a fight.
Still, I dream of the modern realization of component-based software, and it's something I've been thinking a lot about for the past few years.