Build your own Zerg

I recently finished reading Wetware: A Computer in Every Living Cell by Dennis Bray. While the book overall had some issues (I would have preferred more details, less philosophizing) it does bring into sharp focus the concept that the cell is performing a computation. In the case of the cell, the computational substrate is protein molecules, but all the parts you need for a general-purpose computer are present - input sensors, logic elements, and output transducers.
The cell contains a cluster of chemotactic sensors (generally at the front, with additional sensors scattered around the cell membrane), whose function is to sense gradients of certain key chemicals in the surrounding medium. The signal from these sensors is run through a complex mesh of protein synthesis steps. What this mesh is doing is performing a computation in chemical space, using thermal agitation along with a whole bag of chemical tricks such as enzyme regulation and phosphorylation tagging. The end effect of this is to implement computational logic, with AND/OR from other sensors, branching conditionals, memory elements etc. The output can be changes in movement (straight swimming vs. tumbling behavior), synthesis of metabolic or chemically active molecules, or changes in cell constitution or structure. The protein computer is assisted by the stored code of DNA with which it's intricately intertwined.
This makes for three main computing paradigms - electronic computers, neural computers, and protein computers. (Also mechanical computers, quantum computers etc. but on a small scale and not as usefully as electronic computers). Really understanding how protein computers work in all the details is a huge undertaking, but in principle is perhaps not quite as daunting as understanding neural computation. While there is a bewildering complexity to the molecular interactions of the cell, with development of the right tools it is feasible that we could tease out the logic pathways, and it is possible that they are less complex overall than the neural pathways.
If we do gain such an understanding then we open the possibility of manipulating the protein computers for our own purposes. Already we are doing this as much as possible, through drug development, and DNA sequencing and synthesis. However a really fine-grained understanding would allow us to build and manipulate molecular-scale structures and very exactly program and alter cell behavior. The most obvious way to translate 'macro-level' instructions to the molecular level would be through manipulation of the cell's DNA, and the introduction of key signalling molecules. Assuming that we can synthesize and introduce DNA with any arbitrary sequence, then we can imagine constructing all the structures and signalling pathways that we need by starting at the molecular level and building upwards. In this way we can essentially translate from the neural domain (or own goals and objectives) to the electronic domain (the detailed steps and machines needed to synthesize DNA) to the protein domain (the effect of introducing this DNA to a cell). The number of things that we can do once we understand all these steps is limitless.
However I do want to add one cheeky speculation which inspires the title of this post - assuming that we develop a complete understanding of both the neural domain and the protein domain, what if one of the things that we decide to build is a molecular-scale pathway between the two? Instead of using macro-scale machines and electronic computers to synthesize DNA sequences, we bypass the intermediate step and build the structures needed to create a direct neural-to-DNA synthesis. We could imagine providing someone with the the ability to synthesize any arbitrary DNA sequence just by thinking about it.
Evolution would have come full circle - protein computers that build neural computers, that build protein computers. Such a creature could guide it's own biological evolution - a proposition that's both terrifying and inspiring at the same time!