I just thought of a molecular manufacturing system that could probably be built today, and might be very useful, at least in the lab.
When building 100-nm structures out of DNA, the final self-assembly step may require a week. This is OK for a lab demo, but too long for a serious manufacturing process.
It might be possible to speed this up substantially, by keeping the sub-components of DNA in approximately the right orientation for self-assembly as they are being fabricated. How to do this? With a framework built of DNA, of course.
I assume the synthesis of large structures goes something like this: First, use Rothemund staples to build each of the components. Rothemund staples bind to a DNA backbone or "scaffold" (Rothemund's term), folding it up into a structure. This is a relatively fast process (~ 2 hours), since the staples are just a few dozen nucleotides and can diffuse quickly. Next, mix the components together, along with more staples to bind them to each other (the staples may have been built into the sides already). The components, being large (over ten thousand nucleotides), diffuse slowly, and it takes a long time for the correct components to find each other.
If the backbone/scaffolds could be bound to a framework before the staples are added, and the framework held them in proximity to each other, then once the staples folded the backbones into components, the components could self-assemble far more quickly. The position of the components would be mechanically controlled by the framework, greatly increasing their effective concentration.
The framework might be quite large - in fact, it could be too large to self-assemble rapidly. But with the help of another framework - even a smaller framework - a large framework could be built quickly, then used to guide the building of products. This would be true molecular manufacturing, including a nano-building-nano aspect.
If several different components are to be built in parallel on the same framework and then joined, the simplest approach is to use several different backbones. Each backbone will bind with its own staples and not with staples meant for another backbone, just as staples bind to the correct location on a single backbone and not to the many incorrect locations. The ability to use multiple backbones, each making a structure, and then quickly join the structures, implies that shorter backbone strands may be used which will diffuse more quickly to their proper place on the framework.
DNA can be made to unzip as well as zip, by the addition of additional DNA strands that bind to a dangling tail on the strand to be removed. This means that the framework can be physically reconfigured during the fabrication process, and the manufactured parts can be removed from the framework.
When it's time to build products consisting of hundreds or thousands of backbones, it may be difficult to find that many different backbones. But a few dozen backbones might be used in combinatorial pairs to build hundreds of different structures in parallel on the same framework. First, attach several copies of each backbone at various places on the framework (perhaps with multiple attachment points to each backbone), so that the framework holds them apart, and each spatially adjacent pair is different. Second, add staples that bond from one backbone to another; they will only be able to bond to physically adjacent backbones. As each structure takes shape, constraining the topology of each backbone strand, staples may be added that bond from one to another location on the same backbone. Even identical backbones, with different sets of staples attached to form different topologies, can have different dangling strands attached to staple them to different other structures. Thus, there is no obvious limit to how large a product can be built in parallel.
(A staple bonded only on one side, to one backbone, and unable to find an appropriate pair backbone, may be removed by adding an unzipping strand. However, it may be simpler to leave dangling staples in place until the correct staple displaces them by zipping to the backbone. It may also be possible to use staples short enough to float free by themselves unless bonded on both sides. If I understand correctly, the problem of unwanted half-staple attachment must already be solved for Rothemund staples to work at all.)
A sufficiently large framework may be accessible with an optical microscope and conventional cell-manipulation tools or optical tweezers. This would open up whole new vistas of rapid actuation and controlled manufacturing. As a simple example, if it's difficult to build a framework stiff enough to keep unwanted backbone pairs apart, the framework could be physically stretched between optical tweezer beads. Getting a bit more ambitious, a framework attached to several beads could be folded on itself in a variety of ways under computer control, perhaps allowing a single set of structures to be bonded to each other in any of several configurations after the structures are built.
The devil is, of course, in the details. But it seems likely that some variant of this basic framework idea could be developed right away. For starters, take several small backbones, staple them together, build structures by adding more staples, then show that the structures self-assemble into a product far more quickly than they would by mixing separate solutions of the structures. Next, show that the product can be removed (by DNA unzipping) from a previously-stapled framework and the framework (perhaps attached to beads) can be re-used. Then, use the framework to build a larger framework out of multiple backbones. Then, demonstrate that the rigidity of the framework can be controlled to keep identical backbones apart during structure creation but allow their structures to self-assemble once they're built. At that point, you can start imagining what one-micron or even ten-micron product you'd like to build out of DNA.
[Edit: Originally this post referred to a 100-nm box built of DNA taking a week for the final self-assembly step. The DNA box exists, but it's 40 nm, not 100, and it apparently didn't take so long to assemble. William Shih's honeycomb DNA structures do take a week for final assembly, and can reach 100 nm. Several paragraphs of this post have been edited to reflect this.]