Mobile Ad Hoc Networks (MANETs) do not rely on any fixed infrastructure and have to maintain communications through battery-powered nodes in dynamically changing topologies. The classic routing algorithms in the conditions of mobility and the changing quality of links tend to experience frequent route breaks, recidivisms, excessive control overhead, and uneven energy consumption- eventually reducing the duration of the network and worsening the delivery performance. In this paper, an Agentic AI-based Adaptive Routing (AAR) architecture is described where each node is represented as a smart agent, which constantly monitors the local conditions (residual energy, link quality, traffic load, mobility trends) and adjusts its routing behavior on its own. AAR is a mix of (i) energy-conscious next-hop routing, (ii) linklifetime prediction to evade volatile relays, (iii) adaptive control intensity (varying the frequency and extent of routing maintenance to existing dynamics), and (iv) fast local recovery to cut the cost of global recovery. Our AAR experiment is performed on NS-3 simulations with 50/100/150 nodes, AAR always increases lifetime, decreases energy consumption, enhances delivery, and minimizes overheads, and latency in agentic, context-adaptive routing can be reduced to allow stability and energy economy in MANETs.