The Agentic AI Adoption Paradox: Barriers to Enterprise Deployment in the Software Industry
Agentic AI adoption in Indonesia is stalling despite strong executive interest and substantial investment. No empirical study has investigated Agentic AI adoption within Indonesia's software development industry. Existing empirical research has primarily examined Generative AI, with limited attention to the distinct characteristics of autonomous Agentic AI systems. Drawing on UTAUT, this qualitative case study investigates this adoption paradox and extend adoption theory by repositioning ethical concern from a peripheral barrier to an endogenous structural factor. Sixteen semi-structured interviews and one focus group discussion were conducted with practitioners across executive, managerial, and technical tiers using purposive sampling to capture diverse perspectives. Data collection continued until thematic saturation was reached. The data were analysed thematically using the Gioia method, with the assistance of Ligre software to identify and organise emerging patterns and themes. The study identifies three key barriers to Agentic AI deployment in Indonesia. First, output unreliability and qualification limitation hinder its adoption in production environments. Second, Indonesia's relatively low labour costs reduce the economic incentive for investing in Agentic AI infrastructure. Third, while Agentic AI offers significant operational benefits, its capabilities also heighten ethical concerns. We suggest that effective governance and monitoring require Agentic AI to have its ethical guardrails embedded at design time rather than retrofitted post-deployment, have transparent and explainable audit logging, and have its autonomy be calibrated based on its reliability and task risk exposure.