Toward a framework for large action models in accounting: opportunities and challenges in the transition from LLMs to LAMs

Recent advances in artificial intelligence (AI) have driven a shift from passive large language models (LLMs) to large action models (LAMs), which integrate decision-making with autonomous execution (Wang et al., 2025). While LLMs excel in text-based tasks, LAMs bridge the gap between cognitive processing and real-world actions. Advancement in LAMs is crucial in business fields such as accounting and auditing, where AI-driven intelligence has the potential to enhance efficiency, improve accuracy, and assist in decision-making. This paper proposes a conceptual framework for LAMs in accounting automation, incorporating domain expertise and robotic process automation (RPA). We outline key architectural strategies and explain how they are integrated into the overall LAM system design. In addition, we present a simulated prototype that demonstrates how LAMs can automate practical accounting tasks. Furthermore, we examine opportunities in financial automation and compliance, as well as risks related to technical limitations, regulatory challenges, and system integration related to LAMs for future research.