The case for the method
Case study · AI Tools & Strategy · Learning Design
Can AI build a rigorous online course, not just a fast one? The argument here is that it can, on one condition: the standards live inside the process, an agent panel drafts, and the faculty member who will teach the course owns every decision.
The claim
Generative AI can do the tireless first-draft labor of building a course. The open question is whether the result is rigorous or only rapid. The method argues yes, with one condition: rigor cannot be improvised. It has to be built into the process so it happens on every module, whether or not anyone remembers to check.
To test that honestly, the demonstration holds subject knowledge constant. Courses are built in fields I do not teach, so a strong result traces to the method and its embedded standards, not to my own familiarity with the material. It is the harder test, and the more defensible one.
The method
The unit of the work is not any single course. It is the skill that produces them. The skill holds every draft to a fixed checklist of online-quality standards: measurable objectives, outcome alignment, authentic assessment, AI resistance, seat time, Universal Design for Learning, accessibility, interaction, and feedback. The agents draft, the skill enforces, a human finishes. Point it at a new subject and the standards travel with it. The courses are the evidence; the skill is the claim.
The agent panel
Three agents draft under the skill’s direction. They are one part of the method, not the whole of it:
- A subject-matter expert supplies the disciplinary content.
- An instructional designer structures that content against the standards.
- A synthetic student, modeled on the incoming learner, stress-tests the draft from the learner’s side.
The evidence
Every structural decision rests on an established framework, and each one is encoded in the skill so it applies the same way every time:
- Backward design. Outcomes first, then evidence, then instruction, so nothing is included that no outcome claims (Wiggins & McTighe).
- Constructive alignment. Outcome, activity, and assessment are made to cohere, not merely coexist (Biggs).
- OSCQR. SUNY’s open online-course-quality rubric, fifty standards governing structure, clarity, and accessibility.
- Regular and Substantive Interaction. The federal requirement (34 CFR 600.2) for weekly, instructor-initiated presence, designed in from week one.
- Universal Design for Learning. Multiple means of engagement, representation, and expression (CAST), kept distinct from accessibility.
- Authentic assessment. The grade rests on performing the real task, not recall.
One of those strands, authentic assessment, resists AI especially well and carries its own treatment. See Authentic assessment.
Human accountability
The skill never ships a course a human has not reviewed, and that guardrail is the whole point. The agents draft, a named faculty expert confirms the content, and the person who will teach the course owns the result. The failure it exists to prevent is faculty content repackaged without consent or review. The human is never taken out of the loop.
Validation
The real test is a field where my own expertise is not carrying the result:
- Advanced Data Science. The first full test, a graduate course built by the agent panel in a field I do not teach. See the course.
- Light & Lasers. A photonics test, with an industry faculty expert reviewing the physics. See the course.
Applied as a check on my own courses, I built AVC 100 and AVC 248 by hand and ran them through the skill afterward only to confirm alignment and accessibility, the same checks the Dial Your Course builder runs. The tool is quality assurance, not the author.
Grounding
The method rests on documented practice in agentic instructional design and on established learning science, not on preference.
- Rapid AI-Assisted Instructional Design: Using Agentic LLM Tools. Applied Sciences (MDPI). doi.org/10.3390/app16083871Peer-reviewed evidence that orchestrated LLM agents can carry real instructional-design work.
- A sequential agent-based approach to instructional co-design. Springer. SpringerA documented agent-panel method, the pattern the SME, designer, and student agents follow here.
- EDUCAUSE (2026). AI and Course Design: Machines Can Help, but Only Humans Can Teach. EDUCAUSE ReviewThe human-in-the-loop principle the build is organized around.
- Wiggins & McTighe, Understanding by Design; Biggs (constructive alignment); Anderson & Krathwohl (2001), revised Bloom’s.The learning-science backbone: outcomes at the right cognitive level, everything designed back from them.
- Online-quality standards: SUNY OSCQR, Regular and Substantive Interaction (34 CFR 600.2), UDL (CAST), and WCAG 2.1 AA.Standards-aligned by construction rather than audited into shape afterward.