The rapid expansion of e-commerce in developing economies has placed unprecedented pressure on urban logistics networks, particularly regarding the "last-mile" delivery segment. In dense metropolitan areas such as São Paulo, traditional deterministic routing methods frequently fail due to stochastic variables including unpredictable traffic congestion, fragmented delivery points, and restricted infrastructure. This paper proposes a Dynamic Swarm Delivery Model (DSDM) based on Ant Colony Optimization (ACO) to address these inefficiencies. Unlike static Vehicle Routing Problem (VRP) solutions, the proposed model utilizes decentralized, agent-based heuristics to adapt routes in real-time, mimicking biological swarm behaviors to navigate complex urban topologies. The study utilizes OpenStreetMap data and historical traffic patterns from the São Paulo metropolitan region to simulate delivery performance under varying degrees of congestion. Preliminary simulation results indicate that the swarm-based approach reduces average delivery times by approximately 18 percent and decreases fuel consumption by 12 percent compared to standard Dijkstra-based routing algorithms during peak traffic hours. These findings suggest that bio-inspired metaheuristics offer a robust alternative for logistics operators facing the combinatorial complexity of megacity distribution.