Automated Skills Orchestration: Teaching Agentic AI Architecture Through Deterministic Execution Frameworks

The emergence of agentic AI systems—autonomous workflows capable of multi-step reasoning, tool invocation, and self-correction—represents a fundamental shift in how information systems professionals interact with technology. Yet a critical architectural tension underlies all agentic AI systems: large language models are stochastic by design, producing variable outputs through temperature-controlled token sampling, while business processes demand deterministic consistency where identical inputs must yield identical outputs every time. This paper proposes a three-layer skills architecture for teaching IS students to navigate this tension: a Directive layer (Markdown-based standard operating procedures that define intent), an Orchestration layer (LLM-powered decision-making using platforms such as AWS Bedrock), and an Execution layer (deterministic Python scripts that handle precise computation, API interactions, and data processing). Drawing on the practitioner-developed skill pattern pioneered by Nick Saraev, the paper demonstrates how this architecture can be formalized as a pedagogical framework and applied across IS curriculum domains including database management, accounting and financial analysis, business intelligence, and project management. The central argument is that IS programs face a consequential pedagogical choice. They can teach students to use AI tools—a competency with a short half-life, as tools change rapidly and tool usage skills are themselves increasingly automatable. Or they can teach students to architect AI systems—a competency that remains valuable regardless of which specific tools, models, or platforms prevail, because the architectural principles are invariant across the technology landscape. Students who build skills through this framework are not learning to use tools; they are learning to design them. They are not operators of AI systems; they are architects. The skill pattern provides a concrete, assessable pathway for developing this architectural capability, and its portable, self-contained structure makes it a natural foundation for a dedicated course in AI systems architecture within the IS curriculum.