A platform ("walker") moves along a track, past several parts-handling devices, each of which can add a piece to the product - or not, depending on the programming of the system.
This is a robotic, mechanical, digital, programmable system, built out of molecules of DNA. The entire thing is far too small to see with an ordinary microscope.
This is a major breakthrough. It's not the first programmable assembly device based on DNA - Seeman did that a few years ago, with a machine that was built of DNA, programmed by DNA, and could build one of four different DNA strands. But this one is a lot more complex and modular and flexible, and feels a lot more robotic.
Anyone who's claimed nanoscale robotic construction can't work will have to eat their words! The only question now is how fast the robots and products will improve.
Looking forward: If they had a way to (1) recharge the parts-holders in the middle of construction, and if they (2) had a way to connect the parts directly to each other rather than to the platform (without the parts joining to each other prematurely while loading the parts-handlers), and if they (3) could step the growing product past the walker so that the addition zone was always in the right position, then they could, in theory, make products of arbitrary size and sequence.
My recent DNA-binding research proposal describes one hypothetical way to address the first two of these requirements. There are other ways to achieve them as well. The third requirement doesn't seem very difficult compared to what they've already accomplished.
Big hat tip to Michael who brought this to my attention the day it was published!