Real-world deployments of low-cost, peer-to-peer Wireless Mesh Networks (WMNs) for communication in under-served settings are hampered by low throughput capacity and high complexity of network control. We present an autonomic smart-agent based WMN design that organizes a node placement (NP) into dynamic network partitions to promote the WMNs' capacity through increased frequency diversity and advanced network control. Through a custom-built simulation framework that supports agent decision making under real concurrency settings, we show that our smart partitioning technique achieves fast convergence to stable partition sets, relying on local information, under extreme node churn conditions. The design is robust to the average WMN NP density and produces partitions isolated at the physical and link layers with the properties of bounded diameter and node degree, and elected partition control node.