Smart cities thrive not only on advanced technologies but on effective governance that strategically harnesses those technologies. This paper examines how AI agents—digital twins, orchestration bots, decision-support systems, and sensor-fusion models—are operationalized to improve city services across mobility, public safety, infrastructure maintenance, permitting, and environmental management. Using a comparative, evidence-based case analysis of global leaders (e.g., Singapore, Seoul, Dubai, Helsinki, Tallinn, Medellín, Buenos Aires), we synthesize technical documentation and official performance reports to (i) describe common architecture patterns such as sensor and data layers, SAOA-style (Sense-Analyze-Orchestrate-Act/Accountability loops), (ii) map data flows and agent capabilities (real-time inference, multi-agent coordination, human-in-the-loop controls), and (iii) quantify operational outcomes. Reported impacts include 15–25% reductions in peak travel times and double-digit increases in transit reliability, faster emergency triage and response, predictive maintenance that cuts service interruptions and inspection cycle time, and measurable savings in fuel, energy, and CO₂ emissions. Cross-domain findings indicate that integration via unified operations platforms and digital twins, coupled with explicit accountability (audit trails, appeal paths, AI registers), is a stronger predictor of sustained performance than any single algorithmic technique. We conclude with implementation guidance for engineering-led city teams: prioritize interoperable data foundations, codify human oversight at decision points, and measure value via service SLAs and cost-to-serve—turning pilots into durable operating capabilities.