Mr. Anissimov (author of TFA) has either dumbed the science down too much or simply doesn't understand what's going on. I'll try to give a summary of the Nature Nanotechnology paper as clearly and concisely as possible.
First, the researchers made a nanodevice with two slots that could accommodate so-called "DNA cassettes" in a programmable way. The DNA cassettes themselves have free ends that can only bond with complementary DNA. Each of the DNA cassettes has an 'A' end (that can only bond with other A-type molecules) and a 'B' end (I'm simplifying this greatly; 'A' has nothing to do with adenine). The cassettes can be inserted into the two slots with either the 'A' end up or the 'B' end up. So this means there are a total of four states for the device: (1) first slot: A up, B down; second slot: A up, B down; (2) first slot: A down, B up; second slot: A up, B down, etc. The researchers were then able to take four target molecules (one for each of the four programmable states) and show that they bonded to their complementary state. Further, by developing an error-correcting scheme, they were able to get the fidelity of the bonding to 'apparently flawless' levels (quoting FTA, more on this in a sec).
A little more explanation is in order. All of the target molecules have an 'A' and 'B' marker on both ends of their strand. Now, say for example the nanodevice is in state 2: 1A down, 1B up, 2A up, 2B down. The complementary molecule to bind this state would have four markers with 'A' oriented downward and 'B' oriented upward on one end of the strand, and 'A' orented upward and 'B' oriented downward on the other end of the strand. The problem with this is that other target molecules which aren't complementary can still bind. For example, the target for the 1A up, 1B down, 2A down, 2B up would fit equally well into this binding pocket upside down. Also, any of the target molecules can bind with half of the binding pocket, leaving the non-complementary end either dangling or only loosely bound. The researchers get around these two problems using their error-correction scheme. It turns out that the correct target molecules bind more tightly to their complements than the incorrect ones. By heating the devices slightly, the researchers can dissociate the incorrect binding while keeping the correct binding intact. This is, I believe, what was meant by the phrase '100% accuracy.' So, in short, it's still exciting research, at least from my point of view, but no one's moving individual atoms with 100% accuracy or any of the hyper-exaggerated nonsense that I've been reading here.