It's easy to get stuck on the idea that a nanofactory will have to be developed from scratch--that nothing can be done until we know how to pick-and-place atoms with tiny robots, and have built the first tiny robot by as-yet-unknown techniques. Molecular manufacturing researchers tend to focus on the end goal of a diamondoid-building nanofactory, because that is where the highest performance awaits, and because until recently it was necessary to establish that such a thing was possible at all. And they tend to focus on going directly to that stage, perhaps out of habit, and perhaps because they are impatient, and perhaps because they think it will be easiest (as opposed to inventing intermediate technologies). So it is easy for an observer to get the idea that nanofactories will have to be developed all at once.
In fact, the difference between today's capabilities and a diamondoid-building nanofactory can be charted on a number of dimensions, and steps can be taken in any of those dimensions, one at a time. Thus, there are probably thousands of different pathways to a diamondoid nanofactory. In many cases, advances in one dimension will synergize with advances in other dimensions, so it may be more efficient to go straight to a new technology that makes bigger leaps forward in several dimensions at once. But if such a technology can't be found for a particular leap, it can be done in a series of incremental steps.
Here is a partial list of the dimensions:
Precision: Today, a number of lithography tools have a precision of about a nanometer--several atoms wide. The goal is single-atom precision.
Flexibility of molecular construction: Today, chemists can design a wide range of molecules, but only with difficulty. Protein sequences can be built easily, but have been hard to design. The goal of molecular manufacturing is to make a wide variety of shapes and functions by direct design that directly translates to automated processes. Positional mechanosynthesis would of course achieve this goal, but there are other ways. Schafmeister's spiro-linked polymers take intricate, highly-predictable shapes. Rothemund's DNA staples make it easy to design and make large DNA structures. Protein folding, and thus protein design, may be about to get a lot easier.
Manufactured building block size: The goal is to build molecules with designed structure in large enough sizes to implement molecular machines. Bigger building blocks will help in two ways: first, they'll be easier to handle robotically, and second, they'll incorporate more functionality in seamless components.
Feedstock (molecular) inputs: These are molecules built by chemical processes, the first stage inputs. Smaller and simpler feedstock molecules will be cheaper and easier to produce, and will allow the product molecule to have finer features. But smaller feedstock also means more operations to build a given size of manufactured block. And we don't currently have the chemistry to put very small molecules together into engineered 3D structures; mechanical guides, whether special-purpose enzymes or general-purpose robots, will help here.
Product material: There are lots of possibilities, starting with protein (which can be as strong as high-quality plastic), all the way to diamond. Useful product materials may include alumina (sapphire) and silica (which can be deposited by enzymes).
Working environment: The working environment for most protein-based activities is water. The eventual goal is vacuum, because solvent would impose high drag on small fast-moving parts. It's known that some enzymes can work in non-aqueous solvents. Liquid xenon may be a useful "bridge" solvent.
Sub-part assembly: Today, there are several ways to join molecular parts together to make larger parts, but most depend on self-assembly: adding features to each molecule that bind to other molecules when they're shaken together. This is inconvenient, slow, and error-prone--though with DNA tags, and possibly with improved protein folding, it may be useful. Other ways to juxtapose molecular parts include monolayer deposition and dip-pen nanolithography writing. As manufactured building blocks get bigger and machines get smaller and more sensitive, it should become possible to robotically position individual building blocks, allowing direct programmable control of heterogeneous structures at that level.
Actuator performance: Today's nanoscale actuators are either slow (e.g. DNA scissors), or not individually addressable (like pH-activated or light-activated actuators), or both. Better actuator performance is sorely needed. An electrically actuated molecule, connected to wires, would be a good first step; a second step would be to use a number of independently driven actuators in parallel.
Molecular machine capability: Today, researchers are beginning to construct a few molecular machines systems, often borrowing them from biology. The goal is to build engineered systems of machines from a broad and interconnectable set of designs.
Manufacturing machine size: Most nanotech today is done with very large machines. Some is done with sub-millimeter machines (MEMS). Note that chemistry requires reaction vessels which are much larger than molecules, though these are shrinking to MEMS size (microfluidics) in some cases.
It may be possible today to chart a commercially viable course from combinations of today's capabilities aimed at primitive manufacturing (for example, of molecular electronics), to better designs with better products. That would drive molecular manufacturing forward; but it should be clear that even in the absence of a nanofactory-directed effort, progress will continue in each of these dimensions.
The day is rapidly coming closer when a fairly small, easily managed project could reasonably be expected to build a general-purpose molecular manufacturing system (though the first one may or may not use diamondoid).