Metasurface (MTS) comprises an array of metaatoms, each reflecting and inducing a phase shift into the incident wireless signal. We seek the optimal combination of phase shifts across all the meta-atoms to maximize the channel strength from transmitter to receiver. Unlike many existing works that heavily rely on channel state information (CSI), this paper proposes a statistical approach to the phase shift optimization in the absence of CSI, namely blind configuration or zero-order optimization. The main idea is to extract the key features of the wireless environment from the received signal strength (RSS) data via conditional sample mean, with provable performance. Furthermore, as a windfall profit, we show that the proposed blind configuration method has a nontrivial connection to phase retrieval which can be utilized for active sensing. In a nutshell, by configuring a pair of MTSs blindly without channel estimation, we not only enhance the channel strength to facilitate wireless communication, but also enable receiver to localize transmitter. All we need is the RSS data that can be readily measured at receiver. Our algorithm is verified in prototype systems in the 2.6 GHz spectral band. As shown in field tests, the proposed algorithm outperforms the benchmarks (e.g., MUSIC) in the active sensing task, and in the meanwhile raises the signal-to-noise ratio (SNR) significantly by about 10 dB.