Glass
The accessibility overlay for your course content.
Museums protect a canvas with conservation glass: a layer that guards the work and makes it clearly visible to everyone, without altering a single brushstroke. Glass does that for your Canvas® course.

How it works

1 · Scan. Glass reads your course the way assistive technology does — pages, documents, images, video captions — and measures what would fail a student who can't see, hear, or click the way you do.
2 · Propose. Where a fix has one correct answer, Glass drafts it. Where judgment is needed — the wording of a link, the description of an image — Glass writes a draft and shows its evidence.
3 · You approve. Every proposed change waits in a simple review queue: approve, edit, or decline, item by item. Nothing changes in your course without your approval.
4 · Receipts. Every finding, repair, and approval is recorded. What you get is not a promise — it's a before-and-after you can hand to your dean, in the numbers of the scanners your institution already uses.

Measured, not promised

715 → 267
findings across our two pilot courses, before and after repair — measured by a third-party scanner's own re-scan (−63%)
296 → 0
"nondescript link" findings — the vendors' single biggest category — taken to zero on one pilot course
1,016,978
caption words verified word-by-word, with human review where it counts

Numbers from our pilot courses at a California community college, Fall 2026. Your course will differ; the receipts work the same way.

Why now — the AI moment

The accessibility gap for disabled students persisted for one blunt reason: the work was manual. Describing every image, verifying every caption word, rewording every vague link — across thousands of pages and hundreds of hours of lecture video — was never going to happen without staff that most colleges don't have.

Modern AI changed that arithmetic. Glass puts AI to work on the heavy lifting — captions verified word-by-word, image descriptions drafted, repairs proposed — through a verification process we've developed that holds machine output to a measured standard before you ever see it. When a call is genuinely a judgment, it goes to a person. What used to be a semester of unpaid evenings becomes an afternoon of approve buttons.

And the honest limit is built in: regulators themselves have noted that fully automated remediation isn't yet good enough to trust on its own. Glass agrees. That's why AI does the labor here, and people do the judgment — every draft carries its evidence, and every change waits for an instructor's yes.

Why repair, not just report

Accessibility scanners are everywhere. They hand instructors a list of hundreds of findings — and then the list becomes the instructor's problem, without staff, without release time, without an instructional designer. Glass exists for the other half of the job: doing the work, with the instructor holding the approve button. Course content only — Glass's accessibility scans never touch student records, submissions, or grades.

Pilots

Glass is in invited pilots with instructors. If you'd like your course scanned — no cost, no effort beyond adding a reviewer role and clicking approve — get in touch.