Syllabus
Agentic AI and Instructional Design
Subtitle: Human Judgment, Workflow Orchestration, and Learning Experience Design
Credits: 3 graduate credits
Delivery: Fully online; asynchronous core with optional recorded clinics in Modules 4, 8, and 13
Expected effort: 6–8 hours per module
Prerequisite: Prior graduate coursework or professional experience in instructional design, learning technology, curriculum design, or a related field. Programming is not required.
Catalog description
This studio course examines how instructional designers can work with agentic AI systems and design agentic learning experiences without surrendering pedagogical judgment. Students learn to distinguish conversational tools from goal-directed agents, allocate design labor, write task contracts, orchestrate tools and memory, evaluate reliability, preserve learner agency, and establish governance boundaries. The course culminates in a connected two-layer prototype: an agent-supported instructional-design workflow and a learner-facing agentic activity.
Essential question
How can instructional designers delegate, orchestrate, evaluate, and govern AI agents while retaining human authority over learning goals, evidence, ethics, and learner consequences?
Course learning outcomes
By the end of the course, learners will be able to:
- Distinguish chatbots, automations, workflows, agents, and multi-agent systems using observable behavior.
- Decompose instructional-design work and classify decisions as Delegate, Co-do, or Retain.
- Write an agent task contract containing goals, context, tools, memory, constraints, acceptance criteria, and stopping rules.
- Implement an instructional-design workflow with tool use, grounded knowledge, human gates, and trace capture.
- Design a learner-facing agentic activity aligned with a learning outcome and evidence model.
- Evaluate pedagogical alignment, learner agency, cognitive effort, accessibility, equity, privacy, and technical reliability.
- Test and revise prototypes using normal, boundary, and failure cases.
- Produce disclosure, governance, and escalation policies that match the actual workflow.
Learning design commitments
- Process before polish: Final artifacts are assessed together with their traces.
- Pedagogy before autonomy: More automation is not assumed to be better.
- Evidence before impression: Claims about learning must be connected to observable evidence.
- Human gates at consequential decisions: AI may advise; designated people remain accountable.
- Intentional friction: Learners must make meaningful attempts before an agent completes central cognitive work.
- Safe-to-fail practice: Course cases use public, synthetic, or de-identified information.
- Tool durability: Concepts and scoring criteria are vendor-neutral.
Required technology
- Moodle access
- A modern browser and a means of producing documents, diagrams, and a five-minute screen-recorded demonstration
- Access to one approved AI interface or the supplied trace-based fallback cases
No paid personal AI subscription is required. Reference walkthroughs use Codex/Claude-style agents and Moodle/MCP workflows, but equivalent systems are accepted.
Major assessments
| Component | Weight |
|---|---|
| Cumulative trace notebook and formative artifacts | 20% |
| Layer A: Designer-with-Agent workflow | 20% |
| Layer B: Learner-facing agentic activity | 20% |
| Evaluation, red-team, and governance dossier | 20% |
| Integrated capstone and five-minute demonstration | 15% |
| Two substantive peer reviews | 5% |
Module schedule
| Module | Topic | Milestone |
|---|---|---|
| 1 | Orientation: Before the Agent | Baseline task and initial trace |
| 2 | What Makes AI Agentic? | Agenticity audit |
| 3 | Dividing Design Labor | Labor map and risk register |
| 4 | From Prompt to Task Contract | Task contract and ECD map |
| 5 | Anatomy of an Agentic Workflow | Workflow and recovery map |
| 6 | Grounded Research and Content Agents | Validated instructional brief |
| 7 | Evaluation and Trust Calibration | Twelve-case evaluation suite |
| 8 | Studio I | Layer A prototype and peer review |
| 9 | AI as a Learning Partner | Agent role and power map |
| 10 | Pedagogy Before Autonomy | Interaction storyboard |
| 11 | Evidence-Centered Agentic Learning | Layer B ECD blueprint |
| 12 | Online Orchestration and Moodle | Moodle prototype shell |
| 13 | Studio II | Layer B pilot and revision |
| 14 | Integration, Governance, and Disclosure | Governance and integration dossier |
| 15 | Capstone Showcase and Reflective Audit | Connected system and transfer audit |
Participation pattern
Most work is independent and project-based. Required peer interaction occurs in:
- Module 8: Layer A prototype review
- Module 13: Layer B usability and learning-value pilot
- Module 15: Capstone showcase response
Optional live clinics are recorded. Students who do not attend receive the same cases, troubleshooting prompts, and instructor response channel asynchronously.
AI use and disclosure policy
AI use is expected because it is the object of study. Students must:
- disclose systems and capabilities used;
- preserve representative prompts, specifications, outputs, edits, rejections, and overrides;
- distinguish their judgment from system output;
- verify citations and consequential factual claims;
- comply with data, copyright, accessibility, and institutional policies.
Undisclosed AI use, fabricated evidence, or a trace that does not match the submitted artifact is a course-integrity issue.
Data protection
Do not enter student records, grades, credentials, unpublished research-participant data, identifiable workplace information, or confidential course materials into an AI system. Use synthetic cases, public data, or instructor-approved de-identified artifacts. A technically impressive prototype that violates this boundary cannot pass.
Accessibility and fallback
Every required activity has a non-proprietary fallback using supplied transcripts and cases. Videos require captions and transcripts. Diagrams require text descriptions. Learner-facing prototypes must support keyboard use, readable headings, adequate contrast, and a non-agent alternative for accomplishing the essential learning task.
Research separation
Coursework and grades are not research data by default. Any DBR or scholarship-of-teaching study using student work requires separate IRB review, informed consent, and a non-coercive alternative. Refusal to participate cannot affect grades.