David Wiley7 min read

What Can AI+OER Look Like in 2026?

What do AI-powered OER authoring - and the resulting AI-powered OER - look like in 2026? Here’s how I’m creating, reviewing, and publishing Relational Thinking, an interactive open textbook for MIS 340: Introduction to Databases at Marshall University.

My authoring and publication process uses a lot of technology that, just two years ago, was too complicated for many faculty to use. Now in 2026, Claude Cowork and Claude Code do all the technical work for me. And they can do it for you, too. 

Writing the Textbook

I created the first draft of the textbook using a process that anyone can install and use in a tool like Claude Cowork via my openly licensed open-textbook-draft skill. I started by describing why I thought there was a need for another book, including (1) existing books don’t address the impact of AI, (2) existing books are full of examples students don’t care about or relate to, and (3) existing books are written at a reading level that might be difficult for some of my students. I then used Claude to generate a draft Table of Contents for a 15-week course, and then I edited it until I was happy. Then I used Claude to generate the level one headings in each chapter, and I edited these until I was happy. Then I used Claude to generate the level two headings in each chapter, and I edited these until I was happy. With a detailed structure I had reviewed and approved in place, I had Claude draft the text for each section of the book. These were all saved as text files on my hard drive in text (Markdown) format.

I then described the approach to assessment I wanted to use: each chapter should end with four types of assessments: a concept check, apply it, practice drill, and case-study roleplay. These should be created as prompts that students copy and paste into Claude or ChatGPT. The prompts cause the AI tool to engage the student in a structured conversation; ask one question at a time, give feedback on each answer, and offer to go deeper on any topic students want to discuss further. I had Claude generate these assessments at the end of each chapter, and I reviewed and edited them until I was happy. Then, I used Claude to draft binary yes/no rubrics - designed loosely following Google Research’s Adaptive Precise Boolean framework - that students can review before doing the assignments. (A Claude-powered assessment tool I’m developing called Kaiwa, that I’ll write about another time, provides immediate criterion-level feedback and a predicted grade if students choose to use it.)

This initial authoring process took about a day.

Editorial Review and Quality Assurance

I asked Claude to create a knowledge graph that records where each concept is first introduced and where it’s used later. The graph found a couple of cases where concepts were used in an early chapter but not introduced and described until a later chapter. (For example, foreign keys were mentioned in a chapter before the concept of keys had been taught.) I worked with Claude to draft and finalize changes to the affected chapters.

At this point I rather embarrassingly realized that I hadn’t explicitly listed the learning outcomes in each chapter (though they were clearly implied by the initial textbook structure). I worked with Claude to draft these based on the chapter content and edited them until I was happy.

This knowledge graph-based QA and the learning outcomes creation are also part of the open-textbook-draft skill.

Interactive Practice Tools

Finally, as we have worked our way through the textbook this semester, I’ve stayed one week ahead of the students, putting the finishing touches on chapters. In many cases, this has involved using Claude Code to create sophisticated interactive tools students can use to practice and get feedback. (Then I’ve extended them to be usable as summative assessments as well.) Claude Code makes these activities so fast and easy to create that I go from initial concept to a completed tool posted online for students to use (and linked up from the associated textbook chapter) in about an hour. These are 100% browser-based tools, completely free and open, with nothing for students to install or configure.

Publishing, Revising, and Remixing (CI/CD)

I used Claude Code to create a Github repository and sync up the chapter files and tools. Now the files are backed up and version controlled in Github. 

Then I used Claude Code to write a script that can publish the chapters to Pressbooks and added the script to Github. 

Then I configured Github so that every time I make changes to any of the chapter or tool files on my hard drive (because of course I keep tweaking them!) and want to publish them, the following Editorial Review / Quality Assurance process kicks off:

  • Blocking checks: the process checks the chapter’s structure, the learning-outcomes format, that activity prompts formatted as blockquotes, images that exist, links to the tools are valid, and Mermaid diagrams render. If any of these checks fail, the publication process stops and I get notified.
  • Tool checks: the process runs unit tests for the tools, checks that the built tool files match their source, and runs a smoke test that opens every tool page in a headless browser.
  • Reading-level check: re-scores every section the change touches. It gives a warning if the reading level is above grade 10.5 but does not block the publication.
  • Concept-order check: runs a check of new content against the knowledge graph. If this check detects concepts being used before they’re introduced it blocks publication.
  • Preview: a read-only practice run against the live Pressbooks book, which attaches an HTML preview of each changed chapter.

When those tests pass:

  • The interactive tools deploy to GitHub Pages first, so links to the tools are live before the chapters are published with the links to the tools.
  • The script then publishes only the chapters whose content, title or position changed:
    • It converts the markdown to Pressbooks HTML and adds the styled learning-objective and activity boxes.
    • It renders Mermaid diagrams to images, uploads local images, and colors SQL and JSON code.
  • Finally, a drift guard checks whether anyone edited a chapter directly on Pressbooks since the last automatic publication. If so, it refuses to overwrite that chapter and opens a GitHub issue with a description of what changed. (A one-click manual run lets me adopt or discard that edit.)

This editorial review / quality assurance process is fully automated and happens in a minute or two. Then the automated publication happens in a minute or less.

Results

Take a look at Relational Thinking. At the end of Chapter 7, you can find activity 7.0 Query Explorer, one of the interactive practice tools built with Claude Code in about an hour. You can also review the rest of the AI-powered assessments at the end of the chapter. Try cutting and pasting one of the activities into your favorite AI tool.

Authoring, revising, and remixing with AI assistance is an absolute joy. You can draft a complete open textbook replacement in a day. You can create sophisticated interactive learning tools in an hour. Any changes and updates can be instantaneously quality reviewed and automatically published, with all changes tracked.

These course materials are freely available but not openly licensed - because you can’t copyright AI output! Instead, Relational Thinking is in the public domain - affording users permission to engage in all of the 5R activities.

Finally, let me emphasize - these are course materials, not a complete course. The real magic still happens during our active learning exercises when we’re together in the classroom. But these freely available, 5R-able, relevant, interactive materials help everyone get prepared for that time together.  

Continue reading