Four Institutional Jobs, Four Different Failure Costs
Education does not have one AI workflow. It has four, and each one costs something different when the vendor disappears.
Name the job before comparing products. Losing a slide generator costs an evening of rebuilding. Losing the only copy of an accredited course costs a validation cycle.
Verified Status of the Education Stack
Every row below is checked on a schedule, and anything resembling a shutdown is reviewed by a person before the label changes.
| Tool | Job | Status | What it means for your institution |
|---|---|---|---|
| ChatGPT | General assistant | Active | Broadest coverage; check the status history before standardising a cohort on it |
| Claude | Long readings, feedback | Active | Strongest on long documents and structured rubric feedback; see alternatives |
| Gamma AI | Course materials | Active | Outline to teaching deck in minutes; export to PDF at the end of every unit |
| Otter.ai | Lecture capture | Active | Transcripts support accessibility duties; keep alternatives tested |
| Hemingway Editor | Teaching clear writing | Active | Small, stable, and offline-friendly; low risk by design |
| QuillBot | Writing support | Acquired | Still usable, but owned elsewhere now: what happened |
Statuses change. The dated page for each tool is always more current than any list, including this one.
Shutdown Risk Scoreboard
Risk here is not a judgement about teaching quality. It reads three things: who funds the product, how much of the product the free tier is, and how visible recent shipping has been.
The Procurement Gap That Kills Vendors
The single most useful chart in education software is not a feature matrix. It is the gap between how long a purchase takes and how long a startup can wait for it.
Why Education AI Tools Actually Die
Across the tracked cases in this graveyard, education tools fail for four reasons, and only one of them is about the product being weak.
Student Data: The Question That Decides Everything
Every other criterion is negotiable. This one is not. Student records are regulated, and the plan tier you use changes what the vendor may do with what you paste in.
For the policy backdrop, the UNESCO guidance on AI in education and the OECD education directorate both publish current work on classroom AI, data protection, and equity of access.
Seven-Step Institutional AI Audit
Run this once a year and no vendor announcement becomes an emergency.
- Inventory what is actually in use. Include the tools staff pay for personally; that shadow stack is where the data risk hides.
- Mark single points of failure. Any tool holding material that exists nowhere else. That short list is your real exposure.
- Export one full course end to end. Slides, documents, rubrics, transcripts. Time it so you know the true cost of a full backup.
- Confirm the data terms in writing. Ask explicitly whether student input trains models, and archive a dated copy of the answer.
- Check accessibility. Captions, transcripts, keyboard navigation, and screen-reader behaviour, tested rather than assumed from a claim.
- Pilot one alternative per job. Rebuild a single module in a second tool and record what did not transfer cleanly.
- Publish one page of policy. Approved tools, what may never be pasted in, the export schedule, and who to tell when a tool goes quiet.
Five Mistakes Institutions Keep Making
- Leaving the only copy inside the tool. Every painful shutdown story on this site begins here.
- Standardising a cohort on a free tier. If the free tier is the whole product, a price rise is scheduled, not hypothetical.
- Approving a pilot without an exit plan. Write down the fallback before rollout, while nobody is under pressure.
- Treating generated facts as sourced. Assistants invent citations, dates, and quotations with complete confidence.
- Letting AI own an assessment decision. Draft feedback is fine; accountability has to stay with a person.
What This Actually Costs
A working education stack is cheap in cash and expensive in exposure. General assistants sit near twenty dollars per user per month, slide and design tools land around ten to fifteen, and transcription is often free below a monthly minute cap. Institutional agreements lower the per-seat price and raise the switching cost, because the integration work is what you are really buying.
The unbudgeted cost is rebuild time. A validated course trapped in a tool that shuts down is not a subscription problem; it is a semester of staff hours plus, in accredited programmes, another review cycle. An export routine that costs twenty minutes per unit is the cheapest insurance in this category, and it is the one line item nobody writes into the pilot.
Keep Exploring the Graveyard
Education overlaps with several other stacks, and the portability question repeats in each. If you are auditing more broadly, see AI tools for teachers, AI tools for presentations, AI tools for transcription and meeting notes, AI tools for research, AI tools for writing, and the full collection index.
Want the underlying data? Browse every tracked tool from the homepage, compare survivors on the leaderboard, read the latest postmortems, or report a tool you believe is winding down.
Bottom line: pick one tool per job, prefer vendors funded by something other than your free tier, match student data to the plan tier that legally covers it, export every unit into institutional storage, and keep every assessment decision human. Do that and the next shutdown costs you an afternoon rather than a semester.