I believe it is now possible to develop a device with controllable moving parts that can build a copy of its structure from significantly simpler components under external computer control.
Lovers of science fiction or disaster scenarios will be disappointed: the device would have only a few dozen parts, would rely heavily on self-assembly, would have no on-board computer, and would require massive quantities of expensive manufactured molecules to be pumped past it just to make it operate at all.
Basically, all it would do is to sort building blocks out of a mixture, stick them together one at a time in linear sequence, and let the resulting chain fold into the desired structure. But it could, in theory, make any desired structure, perhaps with with low enough error rates and high enough speed to assemble hundreds or even thousands of components. And my design only needs dozens.
Based on what I learned at the recent FNANO conference (which had lots and lots of information about DNA structures, and world-class experts on them), I think all the necessary techniques and knowledge are in place. Several of those experts told me that my idea is interesting; no one told me it was impossible.
The practical utility of this design is near zero. But it would blow away an important conceptual barrier, and demonstrate several useful new techniques and approaches (including one that Eric Drexler proposed years ago).