以场景为牵引的政务大模型应用:范式转型、实践分野与能力生成OA

With the accelerated embedding of generative artificial intelligence in the public sector,the application of government large language models(LLMs)is shifting from technological experi-mentation toward governance practice.The central issue has accordingly moved beyond technical feasibility to how intelligent capabilities can be institutionally absorbed and transformed into gov-ernance capacity.From a scenario-driven analytical perspective,the application of government LLMs can be typologically differentiated into several representative scenarios,including public serv-ice and hotline services,administrative procedures and one-stop government services,internal ad-ministrative operations,and urban governance and platform-based applications.These scenarios ex-hibit substantial variation in governance targets,modes of institutional embedding,and operational logic.Through scenario-based embedding,government LLMs primarily contribute to the gradual formation of governance capacity in areas such as information processing,administrative process op-eration,and the refinement of public service delivery and policy implementation,with their capacity effects displaying processual,structural,and cumulative characteristics.However,government LLMs should not be regarded as universally applicable governance tools.Scenario-driven application represents a governance-oriented pathway centered on institutional stability and endogenous capacity formation,the institutionalization of which depends on the clarification of scenario boundaries,the coordination of rules and procedures,and the sustained development of organizational learning mechanisms.