FROM EMPIRICAL TO DATA-EMPIRICAL AGILITY: DESIGNING AI-AUGMENTED LEARNING SYSTEMS IN AGILE ORGANIZATIONS (Introducing the Data-Empirical Agility Model — DEAM)

Abstract

In the age of artificial intelligence, traditional Agile organizations face new challenges in learning and adaptation. This paper examines how Agile organizations transition from traditional empirical process control toward data-empirical learning under the growing adoption of artificial intelligence tools. The study aims to justify and design a learning architecture in which AI provides continuous sensing signals and recommendations, while humans retain the central role in sense-making, decision-making, and ethical governance. Using a Design Science Research approach, the paper’s main result is the Data-Empirical Agility Model (DEAM), which conceptualizes AI-augmented organizational learning across three interrelated layers: the methodology layer (learning governance and organizational readiness), the framework layer (coordination structures and feedback mechanisms) and the method layer (adaptive practices enacted through continuous improvement cycles). To strengthen practical relevance, the paper provides an exploratory empirical illustration based on anonymized release-level data from five Scrum teams collected across 2025 in a financial domain, including Velocity, Throughput, Lead Time, CRS, defects (bugs) and complementary AI Adoption & Engagement and AI Tools Daily Usage indicators. The observed patterns suggest reduced learning volatility and shorter feedback delays in contexts with higher AI sensing density: across later release cycles, delivery metrics display lower variance between releases, pronounced CRS spikes occur less frequently and late-discovered defects become rarer, even when average throughput remains relatively stable. These findings support the internal coherence of DEAM and indicate that the primary value of AI in Agile learning systems is not merely output acceleration but improved learning stability and more manageable adaptation. Limitations include a single-organization context and the illustrative (non-confirmatory) nature of the empirical component; future research should extend the study through multi-organizational comparisons, quantitative construct validation and examination of moderators such as AI trust, psychological safety and governance policies.