WFIRST will conduct a coronagraphic program of characterizing the atmospheres of planets around bright nearby stars. When observed with the WFIRST Wide Field Camera, these stars will saturate the detector and produce very strong diffraction spikes. In this paper, we forecast the astrometric precision that WFIRST can achieve by centering on the diffraction spikes of highly saturated stars. This measurement principle is strongly facilitated by the WFIRST H4RG detectors, which confine excess charges within the potential well of saturated pixels. By adopting a simplified analytical model of the diffraction spike caused by a single support strut obscuring the telescope aperture, integrated over the WFIRST pixel size, we predict the performance of this approach with the Fisher-matrix formalism. We discuss the validity of the model and find that 10 µas astrometric precision is achievable with a single 100 s exposure of a R AB = 6 or a J AB = 5 star. We discuss observational limitations from the optical distortion correction and pixel-level artifacts, which need to be calibrated at the level of 10 -20 µas so as to not dominate the error budget. To suppress those systematics, we suggest a series of short exposures, dithered by at least several hundred pixels, to reach an effective per-visit astrometric precision of better than 10 µas. If this can be achieved, a dedicated WFIRST GO program will be able to detect Earth-mass exoplanets with orbital periods of 1 yr around stars within a few pc as well as Neptunelike planets with shorter periods or around more massive or distant stars. Such a program will also enable mass measurements of many anticipated direct-imaging exoplanet targets of the WFIRST coronagraph and a "starshade" occulter.