Designing Adaptive Enterprise Workflow Architectures with Generative AI and Retrieval-Augmented Generation

Abstract

Enterprise workflows are usually designed once and changed rarely, while the conditions they operate in change constantly. Regulations are revised, volumes shift, bottlenecks appear, and new kinds of cases arrive. Configurable workflow architectures made change easier by separating process logic from code, but deciding what to change and when still depends on slow, manual analysis. This article proposes an architecture for adaptive enterprise workflows in which generative artificial intelligence (AI) and retrieval-augmented generation (RAG) close the loop between observing a process and improving it. Building on the Monitor–Analyze–Plan–Execute over a shared Knowledge base (MAPE-K) model of self-adaptive systems, the architecture uses execution monitoring and process mining to detect problems. RAG-grounded language models diagnose causes and draft adaptations from policies and past changes. Bounded autonomous agents carry out approved changes through a configurable workflow engine. Guardrails and an audit-ready compliance layer surround the whole loop. The article defines three levels of adaptation, from individual case routing to structural redesign, each with its own approval and evidence requirements. It illustrates the architecture with a bank's customer onboarding workflow responding to a regulatory change and a processing bottleneck. It argues that workflows can safely become adaptive only when AI autonomy is matched to the scope of each change and every adaptation is traceable.