A Reference Architecture for Integrating Agentic AI into Enterprise Software: Responsibilities, Interactions, and Variability

This reference architecture decomposes enterprise agentic AI into ten distinct functional responsibilities, including orchestration, memory, and tool execution. It defines technology-neutral boundary invariants and typed cross-responsibility interactions to standardize how reasoning, enterprise context, and external actions integrate into existing software stacks.