Models and Concepts of AI Agents in Financial Operations for Autonomous Payroll Processing

The study examines models and architectural concepts of AI agents embedded in financial operations to support autonomous payroll processing in multi-region, compliance-sensitive environments. The research focus lies on agentic patterns that orchestrate payroll data flows end-to-end over ACID-compliant lakehouse platforms, feature stores, and metadata-driven orchestration layers. The work synthesizes current literature on AI agents in finance, autonomous decision systems, data lakehouse architectures, and feature-store-centric machine learning pipelines, and combines it with design patterns emerging in production-grade payroll and HR systems. Particular attention is given to deterministic computation graphs, idempotent pipelines, concurrency-safe merge patterns, and real-time observability for reconciliation and audit. The article aims to formulate a conceptual model of an autonomous payroll stack, outline classes of domain-specific agents, and identify their limitations and governance needs. The material is intended for researchers and practitioners in AI, data engineering, and financial systems design.