Doesn't nature have a lot of inherent stochasticity down at the molecular level that makes it very difficult to do positional, precise molecular manufacturing, due to everything jittering around randomly?
Actually assembling individual atoms together to make stuff is still very much an unsolved problem. Freitas' work on mechanosynthesis tooltips shows that we may need to control more than just position to precisely react atoms and that 'fat fingers' is still very much a problem.
https://www.youtube.com/watch?v=Ms_ehUVvKKk
How far could we go by modifying existing biological molecular machines? Probably quite far.