AI-assisted production workflow for scalable course builds
Two programmes, ten Canvas courses, and three months to manage the digital build.
As the sole Learning Technologist, I needed to build seven courses while coordinating and overseeing a vendor responsible for the remaining three.
I created a connected set of AI agents covering the production process from storyboard review to Canvas build and build review. Testing across two courses increased my capacity from approximately one module per day to three, while improving the accuracy and consistency of the digital builds.
Project overview
FourthRev needed to produce two online programmes for a leading Australian university between July and October 2026.
I was responsible for building five courses in one programme and two in the other, while coordinating the vendor building the remaining three.
A conventional manual process would not have been sustainable within the timeframe. Using Claude Cowork and a Chrome extension connected to Canvas, I built a set of AI agents to review storyboards, create Canvas content, manage media and interactives, and check the completed builds.
The result is an end-to-end production system designed around branding, accessibility, build accuracy, and maintainability.
Key features & early outcomes
- Built a connected set of AI agents covering storyboard review, Canvas production, interactives, media, and build review
- Created the global CSS, JavaScript, and matching Canvas markup for the university sub-account
- Automated the creation of pages, quizzes, and assessments from approved Google Workspace storyboards
- Established a maintainable workflow for GitHub-hosted interactives embedded in Canvas
- Tested and refined the system across two courses
- Increased production capacity from approximately one module per day to three, with more accurate and consistent builds
As the sole Learning Technologist, I needed to manage the digital production of ten courses within three months.
My scope included building seven courses directly and coordinating the vendor responsible for the other three. Across both programmes, I also needed to maintain consistent branding, accessibility, media handling, assessment setup, and build standards.
A manual page-by-page process would have consumed too much time in repetitive production. I needed a system that could handle this work reliably while leaving me in control of structure, exceptions, build decisions, and release readiness.
I designed a connected production workflow in which each agent handles a defined stage of the digital build.
The workflow runs through Claude Cowork, using the Chrome extension to work directly in Canvas LMS and the Google Drive connector to retrieve the approved storyboard. Each agent is configured as a Claude Skill with clear instructions defining its task, production rules, and expected output.
1. Canvas design system
I used AI-assisted coding to create the global CSS and JavaScript for the Canvas sub-account, along with reusable HTML markup patterns.
This gave every course a consistent visual and technical foundation before production began.
2. Storyboard-review agent
This agent checks whether a storyboard is ready to be converted into Canvas.
It identifies missing information, unclear structure, and incomplete build instructions before they create problems in production.
3. Canvas build agent
The build agent retrieves an approved storyboard from Google Workspace and converts it into Canvas content through the Chrome connection.
It applies the established markup and creates pages, quizzes, and assessments directly in the LMS.
4. Interactive-development agent
For interactions beyond standard Canvas components, this agent creates the interactive, publishes the code to GitHub, and embeds it in Canvas.
This makes the interactives maintainable and reusable while keeping them visually consistent with the course.
5. Media agent
The media agent handles Wistia and H5P embeds, image uploads, and media placement.
It applies the storyboard instructions and production rules consistently across the courses.
6. Build-review agent
The final agent compares the completed Canvas build with the approved storyboard and build standards.
It checks whether the digital conversion was completed accurately, including structure, formatting, media, embeds, interactions, assessments, accessibility, links, and missing content.
This is separate from editorial or subject-matter review. Its purpose is to catch issues introduced during the digital build before formal QA and release.
A connected system creates the real efficiency. The largest productivity gain came from connecting storyboard review, Canvas production, media, interactives, and build review — not from automating isolated tasks.
Storyboard clarity determines build accuracy. Testing the agents revealed where storyboard instructions were incomplete or ambiguous. Refining the source improved the Canvas builds and made subsequent production more reliable.
Automation needs a stable design system. The global CSS, JavaScript, and markup patterns gave the agents clear production rules and kept the courses consistent.
Build review can improve the whole workflow. When the build-review agent identifies a recurring conversion issue, I can update the relevant agent so the correction carries into later courses.
AI can significantly expand an individual role. The workflow increased my capacity from approximately one module per day to three, making it possible to manage a much larger production scope while maintaining build accuracy and consistency.