Maximum sensitivity for optical interferometers is achieved only when the optical pathlengths between the different arms can be equalized without using interference fringes on the research object itself. This is called “blind operation” of the interferometer. In this paper I examine different options to achieve this, focussing on the application to the Very Large Telescope Interferometer (VLTI). It is proposed that blind operation should be done using a so-called “coherence autoguider”, working on an unresolved star of magnitude V=11 to 13 within the “isoplanatic patch for coherencing”, which has a diameter of about 1 degree. Estimates of limiting magnitudes for the VLTI are also derived.