Early-stage real-estate development decisions bring together planning and regulatory constraints, spatial feasibility, market positioning, cost and schedule logic, and financial risk under uncertainty. In the Italian/EU context, this complexity is reinforced by demanding approval processes, fragmented local data, and strong expectations for documentation and accountability. Recent advances in artificial intelligence (AI), especially large language models (LLMs), suggest a faster way to organize feasibility work and compare alternative options. However, practical use is still limited by hallucinations, inconsistent outputs across iterations, privacy concerns, and the difficulty of producing spatially verifiable results. This thesis investigates to what extent AI can support an end-to-end early-stage feasibility and decision-making process for a complex urban redevelopment project in Italy, and which governance mechanisms are required to make the outputs reliable, auditable, and usable in practice.