doi.org
AI Academy Session 4: Agentic AI — From Asking Questions to Delegating Tasks
Instructional materials for the GW Engineering AI Academy, exploring the shift from conversational AI to "agentic" workflows, where AI systems act with increasing autonomy to execute multi-step tasks.Key topics covered include:The Anatomy of an AI Agent: Defining the "thought–action–observation" loop and the three essential ingredients for agency: reasoning, tools, and memory. The Model Context Protocol (MCP): An explanation of the industry-standard protocol that allows AI models to connect seamlessly with external services like Google Drive, Gmail, and GitHub. Agent Skills: Introduction to a standardized format (SKILL.md) for packaging domain-specific expertise—such as ABET accreditation mapping—into portable, reusable instructions for AI agents. Practical Workflows: Strategies for faculty to transition from manual prompting to orchestrating autonomous assistants that can analyze course outcomes, review student reports, or conduct deep research. The document includes an appendix demonstrating a concrete example of an agent skill for mapping Course Learning Outcomes (CLOs) to ABET Student Outcomes. See the full skill at: https://github.com/labarba/abet-clo-mapper GW AI Academy, session 4, April 3, 2026.
