As data grow and become more complex, organizations face increasing challenges in automating and scaling analytics. Conventional analytics tools often prescribe static workflows with limited adaptability and usually require manual intervention in several stages of the process. This thesis puts forth AIDST, the AI Data Science Team, a multi-agent system architecture for autonomous running of end-to-end data analytics workflows. According to Design Science Research (DSR) methodology, this work proposes a modular framework of intelligent agents arranged into perception, cognitive, and execution layers. These agents cooperate to ingest data, engineer features, train models, and deliver results through a well-specified protocol for inter-agent communication. Implemented with Python, MLflow, Streamlit, Docker, and Kubernetes, AIDST is validated through benchmarking and domain-specific case studies across finance, healthcare, and manufacturing. Results demonstrate a 14% reduction in the MSE for regression problems, a 6.7 percent lift in classification accuracy, a 22 percent reduction in runtime with a 3.7× speed-up in workflow execution, and 99.4 percent uptime for 72 hours with 89 percent automated conflict resolution. This means that, in all respects, AIDST gained scalability, adaptability, and in usability over existing approaches. This work advances the emerging field of agentic AI systems by presenting a scalable paradigm for intelligent autonomous data science operations.