Software-defined vehicles (SDVs) are shifting automotive engineering from fixed electronic-control-unit deployments toward continuously evolving cloud-connected, edge-compute, and AI-assisted software platforms. This transition enables adaptive features, predictive diagnostics, cybersecurity monitoring, over-the-air software updates, and data-driven user experiences; however, it also increases architectural complexity, update risk, attack surface, and operational uncertainty. This article proposes an AI-enabled cloud-edge architecture for adaptive, secure, and resilient automotive software systems. The architecture integrates centralized and zonal compute domains, Android-based embedded application services, secure cloud-to-edge update orchestration, edge AI inference for anomaly and risk scoring, zero-trust access enforcement, software-bill-of-materials-aware package validation, and recovery mechanisms for interrupted update and application workflows. Unlike conventional SDV reference models that separately discuss AI, OTA updates, cybersecurity, and resilience, the proposed architecture treats them as a unified lifecycle problem: software must be selected, validated, deployed, monitored, recovered, and improved across vehicle, edge, and cloud layers. The paper defines the system layers, data/control flows, AI model placement strategy, threat model, update workflow, resilience state machine, and evaluation metrics. The proposed framework is aligned with automotive software-update engineering, cybersecurity engineering, secure software development, AI risk-management, and cloud-edge security practices available through March 2024. The result is a practical technical architecture intended for next-generation automotive and embedded mobility platforms requiring reliable software evolution under safety, security, latency, and resource constraints.