Smart dust is a term invented several years ago at Berkeley. I probably heard it for the first time in the 90ies from prof. Sangiovanni Vincentelli who in his lab was trying to miniaturise motes, sensors with communications capability. At that time a mote was like a cubic cm, amazingly small for the computers we were used to have at that time. At the turn of the century it shrunk in the region of 1 cubic mm.
I remember he telling me that in the future their dimension will shrink to the point of becoming almost invisible and the economics would allow the manufacturing of thousands of them. The main challenges were the powering and the coordination of the ... swarm.
Fast forward to the next decade.
The problem of powering the motes will be solved, thanks to electronics becoming less power hungry and to the increase effectiveness of scavenging (that is the capability to harvest power from the ambient, like converting like converting light, sound, vibration, temperature, chemical energy into electrical power).
This leaves the problem of orchestrating the hundreds thousands of motes to create an effective swarm delivering the desired functionality.
Part of the orchestration (like the communication part among the motes) has already been addressed and by the next decade we can assume it will be taken for granted. Take a look at the second clip showing the orchestration of 500 drones by Intel, in November 2016. They already showed the orchestration of 100 drones earlier in 2015 but moving up to 500 was a completely new challenge. Imagine moving up to thousands....
What remains is the need to "program" the swarm to perform specific functionalities. You can imagine being asked by a farmer to spray the dust on a field to check for the emergence of certain bacterial (or insect) contamination. You will have to pick up the right kind of dust (with the right sensors) and then program the swarm to detect the bacteria by monitoring certain data that all together can be mined to understand what is going on. Smart dust swarms and big data will go hand in hand, since you must keep the functionality of the single grain of dust as simple as possible to go easy on power and then use the data flowing from thousands of them to understand what is going on.
There will be the need for skill in the application domain (like farming and biology in this example) as well as in data analytics and system wide coordination.
The fields of application will just keep growing. Smart dust will be a sort of third millennium screwdriver required skilled hand to use it.