The pressure
Forty courses a year. Too much rework. AI is already in the conversation.
An education company produces 40 custom courses per year. They want AI's speed without giving up quality, privacy, or IP control.
- Inconsistent SME inputs
- Repetitive restructuring
- Long review cycles
- Weak traceability
- Interest in AI
- Concerns about privacy, hallucination, IP, and quality
The operating change
The workflow does not become “AI-first.” It becomes explicit.
- Client brief
- SME documents
- Manual analysis
- Outline
- Storyboard
- Development
- SME review
- Revision
- QA
- Release
- Structured intakeHuman
- AI-assisted analysisAI
- Human validationHuman
- AI-assisted outlineAI
- ID reviewHuman
- Controlled draftingAI
- SME validationHuman
- Automated QAAI
- Human releaseHuman
Your turn · Nine production decisions
Would you let AI do this?
Classify each task, then compare your judgment with mine. The disagreements are more useful than a generic list of AI use cases.
01 Summarize source material
02 Draft learning objectives
03 Find missing information
04 Draft assessment items
05 Approve technical accuracy
06 Choose instructional strategy
07 Check consistency
08 Publish to LMS
09 Evaluate learner performance
The non-negotiables
Seven gates AI never passes alone.
Not “someone reviews it.” A named person checks a named thing before the work moves forward.
Source grounding
Every AI claim traces to an approved source document.
Privacy review
No confidential or personal data enters an ungoverned model.
SME approval
A qualified expert signs off on technical accuracy.
Instructional review
A designer confirms strategy, alignment, and objectives.
Accessibility review
Output meets accessibility standards before release.
Bias review
Content is checked for biased or exclusionary language.
Release approval
A named human authorizes publication.
What the design changes
The operating model, before and after.
| Measure | Before | After | Why |
|---|---|---|---|
| Intake | Every SME sends something different | Required fields and approved sources | AI cannot repair an intake process nobody has defined. |
| Drafting | Writers restructure and rewrite manually | AI drafts inside source and template constraints | Acceleration happens inside a bounded task. |
| Review | Responsibility shifts between people | Named technical, instructional, accessibility, and release owners | A person—not “the team”—owns every consequential check. |
| Traceability | Corrections are hard to trace | Sources and approvals travel with the asset | The workflow preserves the reason an output was accepted. |
These projects are original solution-architecture demonstrations created to show my consulting approach. The organizations, requirements, data, budgets, vendors, timelines, and results shown are fictional and do not represent previous clients or employers.