The alternative to constructing atomically precise manipulators with other atomically precise manipulators is to construct them with macroscopic machinery such as scanning tunneling microscopes. Like writing a compiler in machine code with front-panel switches or copying an operating-system kernel with a handheld hole punch and paper tape, this may be feasible, but it will be extremely difficult; once we have a working atomically precise manipulator, it will be an enormously easier way to build more of them with it.
That's the reason for Drexler's prudent concern for safety protocols in advance of the actual development of the machinery.
Of course, there's no reason that the constructor itself need be a micron-scale device; the convergent-assembly proposals from Zyvex and Nanorex instead use macroscopic machines with a large number of nanometer-scale atomically-precise parts.
We need to clean this place up.