Presentation by Jon Fila, Inclusive AI Strategies. Use the arrow keys, Page Up and Page Down to move through slides and step reveals. Press M to turn motion off or on, F for fullscreen.

Idea to Pilot-Ready in Under a Week

Think Bigger About AI

Jon Fila

Founder and AI Strategist, Inclusive AI Strategies™

QR codeGet these slides iais.us/ideadlac26 (opens in a new tab)

Beside the title, a grid of course modules: six units with yellow caps, five rows of green cards in each.

Where AI training stops

  • LESSON PLANS

    Generate a lesson in the style of…

  • QUIZZES

    Write five multiple-choice questions…

  • PASSAGES

    Rewrite this at a 9th-grade level…

Valuable as a starting point. Being treated as the destination.

Task level vs. system level

The course grid is empty: six units, five slots each.

TASK LEVEL

One artifact. One prompt. One reviewer.

SYSTEM LEVEL

A full course, aligned to standards.

UDLSELWCAGbuilt into every cell

Building one online course

Three empty course grids side by side, one for each era.

Over a decade ago, by hand

800+

hours

Two years ago, with early AI tools

350+

hours

Now, about

one week

 

AI didn’t build it. I did.

Course wish list

What I have always tried to build into a course

  • UDL principles
  • SEL frameworks
  • AI guidance and AI-resistant assignments
  • SPED accommodations and ELL support
  • Common Core alignment
  • WCAG compliance
  • Instructor slideshows, guidance and support
  • ISTE Standards
  • NSQ Online Course Standards
  • Depth of Knowledge
  • Inclusive design
  • Community of Inquiry

THE COURSE

Creative Writing Through Dungeons & Dragons

Students want to tell stories.

  1. UNIT 1

    World Building

  2. UNIT 2

    Character Development

  3. UNIT 3

    Narrative Structure

  4. UNIT 4

    Dialogue and Voice

  5. UNIT 5

    Campaign Design

  6. UNIT 6

    Collaborative Storytelling and Capstone

  • Grades 6–12
  • Common Core ELA
  • NSQ aligned
  • Built to WCAG 2.2 AA
  • CC BY-NC 4.0

Free course: iais.us/dndcourse

Course details from inclusiveaistrategies.com/dnd-writing-course and the course’s unit pages. Dungeons & Dragons is a trademark of Wizards of the Coast. This course is not affiliated with, endorsed by, or sponsored by Wizards of the Coast.

How I designed the course to make AI shortcuts unappealing

  1. Personal connection

    Every major assignment is rooted in the student’s own place, memory, or voice.

  2. Metacognition

    “Why did you choose this?” is embedded in every assignment.

  3. Compounding work

    The character from the backstory returns later. AI can’t fake continuity it didn’t make.

  4. Capstone synthesis

    The final project weaves together the learner’s own creative work from across the course.

AI GUIDANCE, WRITTEN INTO THE ASSIGNMENT

What students are told they may do

You may ask an AI tool to ask you questions about your hometown to help you remember details.

Assignment 1.3: Hometown Fantasy Setting, Creative Writing Through Dungeons & Dragons

Built for WCAG 2.2 AA

VISUAL

  • Atkinson Hyperlegible font
  • WCAG 2.2 AA color contrast
  • 1.5 line height
  • High-contrast callout boxes

STRUCTURAL

  • Consistent heading hierarchy
  • Tables with proper headers
  • Predictable “Next Steps” navigation
  • Logical reading order

CONTENT

  • Alt text on all visuals
  • Descriptive link text
  • Plain language with jargon defined
  • Multiple formats for key concepts

Applied consistently across every HTML page in the course.

Standard: Web Content Accessibility Guidelines (WCAG) 2.2, W3C Recommendation, w3.org/TR/WCAG22

What changed, and what stayed constant

One finished course grid.

  • TIME800 hours to 350 hours to about one week
  • SCALEOne course to courses in parallel
  • REVIEW DEPTHOne pair of eyes to multi-frame audit
  • ITERATION COSTHigh to near zero
  • ACCESSIBILITYRetrofit to designed in
  • An instructional designer reviews every piece.
  • Standards shape the course from the first outline.
  • Quality checks happen at every stage.
  • Accessibility and equity are planned from the start.
  • People who know the learners make the calls.

AI doesn’t replace expertise. It operates at the speed of it.

How AI use may affect experts and novices

EXPERTS

AI helps most at the practice stage, for students with strong content knowledge

As a tutor, study partner or editor

NOVICES

When AI does the thinking, all students risk skill decay

  • With AI50%
  • Without AI67%

Developers learning a new coding library: follow-up quiz scores (2026)

Sources: Kendeou, Greene, Nixon et al., “Ways Generative AI Can Support and Threaten Learning in K-20 US Education: Expert Consensus Report,” University of Minnesota, October 2026; Shen and Tamkin, “How AI Impacts Skill Formation,” arXiv:2601.20245, 2026 (preprint).

What AI won’t do by default

  • Choose the framework

    NSQ, QM, UDL, WCAG: which standards, in what order, for what learner?

  • Make equity calls

    What counts as culturally responsive in this community, with these texts?

  • Sequence cognitive load

    What does a learner hold in working memory in week three?

  • Know your learners

    IEPs, ELL supports, the student who just lost a parent. Context is not in training data.

Check the work in passes

  1. Generate.

    Treat the first output as a draft.

  2. Check mechanically.

    Run the tools that measure: accessibility and contrast checkers, link checks, standards alignment against the actual documents.

  3. Review adversarially.

    Ask a second model, or a colleague, to look for errors. Compare the two reviews.

  4. Read as the learner and the expert.

    Open it the way a student will, and read it the way a subject expert will. Automated checks cannot do this pass.

The slides for this talk were built with AI and an explicit WCAG 2.2 AA requirement. They were reported as compliant. A rendered contrast audit still found failures, including bullet markers that were nearly invisible.

Designed in, not applied after

APPLIED AFTER

  • Build the course first
  • Audit for accessibility at the end
  • Bolt standards labels onto existing content
  • Retrofit equity where it was thinnest
  • Continual iteration and review

DESIGNED IN

  • Standards drive the course architecture
  • WCAG checks built into every template
  • NSQ criteria reviewed at each module
  • Quality rubrics considered from outline to final
  • Quality by default, not by rescue

Testing Claude on a course team’s quiz review

A fall 2026 cohort of online course designers. One dot per run.

  • Caught all six planted quiz problems9 of 10 runs
  • Applied the team’s question-bank depth rule without being told0 of 4 runs
  • Caught the platform’s 150-character field limit0 of 6 runs
  • Named the team’s approved accessible alternative0 of 6 runs
  • Once given the team’s rules: every run flagged all six activity problems3 of 3 runs

A detailed step-by-step prompt caught no more than “Check these quiz items for problems.” Adding the team’s rules is what changed the results.

Source: Jon Fila, pre-test built for the cohort, October 2026. Claude Sonnet run from the command line with a minimal system prompt; 3 to 10 runs per check; scored against an answer key and checked by hand. Results in the chat app may differ.

The training gap

TASK-LEVEL PD

What most AI PD teaches

  • Prompting basics
  • Tool selection (ChatGPT vs. Claude)
  • Quick artifact generation
  • Detection and AI-integrity policies

Prepares educators for individual classroom use

SYSTEM-LEVEL PD

What full-course AI design needs

  • Layered context: standards, local rules, platform limits
  • Standards-embedded workflows
  • Multi-frame review (accessibility, equity, craft)
  • Governance and QA at course scale

Prepares educators for course and program design

Three barriers to system-level work

The tools are ready for system-level work. What’s stopping institutions from getting there?

  1. Expertise depth

    Institutions assume AI replaces instructional design expertise. It doesn’t. It requires more of it, earlier in the process.

  2. Quality framework

    Without standards designed into the workflow, AI reproduces the field’s default mediocrity, faster.

  3. PD scope mismatch

    Task-level training cannot prepare educators for system-level course design. We need a different PD.

Three concrete next steps

  1. Audit your current AI PD

    Task-level or system-level? If it’s all task, name it and plan the next layer.

  2. Map one course to quality standards

    Pick one. NSQ, QM, WCAG at the outline stage. Feel where the work actually is.

  3. Identify one system-level pilot

    One course. One quarter. System-level from day one. Measure against task-level comparables.

Effective and efficient LLM use

  • Start with a profile

    Who are you? The tool? The audience?

  • Work in persistent project spaces

    Add files and instructions to a project so everything relevant is in context when you begin.

  • End long sessions with a handoff

    Have the tool create a handoff.md to begin the next session.

  • Keep a corrections list

    Have the tool create a lessons.md to learn from corrections.

  • Build a reusable prompt library

    Use the tool to refine your prompts.

  • Specify output shape before you prompt

    Know what an ideal finished product looks like.

Handout: iais.us/aiworkflow

WHAT CUSTOMIZING MEANS HERE

The Default Sucks.

  • FRAMEWORKLoad your standards and framework before the first prompt.
  • LOCAL RULESGive AI your own rules. It won’t find them on its own.
  • DESIGNBuild accessibility and equity in from the outline.

Questions?

QR codeGet these slides iais.us/ideadlac26 (opens in a new tab)
Jon Fila in a navy blazer and white shirt against a gray background

Jon Fila

Founder and AI Strategist, Inclusive AI Strategies

  • jon@iais.us
  • inclusiveaistrategies.com
  • linkedin.com/in/jon-fila

Sign up for my newsletter

inclusiveaistrategies.com/#newsletter