Using wide field narrow-band surveys, we provide a new measurement of the $z=6.6$ Lyman-$\alpha$ Emitter (LAE) luminosity function (LF), which constraints the bright end for the first time. We use a combination of archival narrow-band NB921 data in UDS and new NB921 measurements in SA22 and COSMOS/UltraVISTA, all observed with the Subaru telescope, with a total area of $\sim 5$ deg$^2$. We exclude lower redshift interlopers by using broad-band optical and near-infrared photometry and also exclude three supernovae with data split over multiple epochs. We spectroscopically confirm the two most luminous Ly$\alpha$ emitters ever found at $z=6.604$ and $6.541$ in the COSMOS field using Keck/DEIMOS and VLT/FORS2. Combining the UDS and COSMOS samples we find no evolution of the bright end of the Ly$\alpha$ LF between $z=5.7$ and $6.6$, which is supported by spectroscopic follow-up, and conclude that \emph{Himiko}-like sources are not as rare as previously thought, with number densities of $\sim 1.5\times10^{-5}$ Mpc$^{-3}$. Combined with our wide-field SA22 measurements, our results indicate a non-Schechter-like bright end of the LF at $z=6.6$ and a different evolution of \emph{observed} faint and bright LAEs. This differential evolution was not addressed in previous studies, or discarded as cosmic variance, but we argue instead that it may be an effect of re-ionisation. Using a toy-model, we show that such differential evolution of the LF is expected, since brighter sources are able to ionise their surroundings earlier, such that Ly$\alpha$ photons are able to escape. Our targets are excellent candidates for detailed follow-up studies and provide the possibility to give a unique view on the earliest stages in the formation of galaxies and re-ionisation process.