This fall, two of the biggest names in AI told schools the same thing: the custom assistants you built on our platform are going away, so rebuild them in our new format. It’s the best argument I’ve seen this year for a district-owned AI platform.
Google is moving Gemini Gems to Skills. OpenAI is retiring Custom GPTs in favor of plugins. Neither change is a scandal. Products evolve. But if your teachers spent the last year building Gems for lesson planning or Custom GPTs for feedback, they just learned an expensive lesson about renting a platform instead of owning one.
I’ve watched this movie before. When I started partnering with schools on 1:1 laptop rollouts back in 2009, the lesson wasn’t about the device. It was that owning and supporting the platform mattered more than whatever shiny thing sat on top of it. Tricia Friedman and I talked about how fast school AI work is maturing on the Shifting Schools podcast episode AI in Schools Is Changing: What Leaders Need Now, and this is exactly the kind of change leaders need a plan for.
Key takeaways
- Google says education accounts won’t lose access to Gemini Gems before June 1, 2027; Gems move into Settings starting November 17, 2026 and are later converted to draft Skills.
- OpenAI is replacing Custom GPTs with plugins; conversations, the selected model, sharing settings and custom actions don’t carry over automatically.
- Every migration costs schools staff time: rebuilding, re-testing, re-sharing and re-training.
- When the district owns its AI platform, the district decides when assistants change, which models they use and who can see them.
What actually changed this fall
Google: Gems become Skills. According to Google’s own admin guide to the transition, Skills rolled out in Workspace starting October 5 and in the Gemini app on October 13. On November 17, Gems move into Settings in the Gemini app. Education editions lose Gem access no sooner than June 1, 2027, and that includes Gems used in Google Classroom and in learning management systems connected through Gemini. Migrated Gems show up as inactive drafts that creators must review and activate before using or sharing. Google also notes that Skills currently don’t sync between the Gemini app and Workspace apps, so a Skill you want in both places has to be built twice. Android Central reported the same timeline.
OpenAI: Custom GPTs become plugins. OpenAI announced on September 11 that plugins, which bundle instructions, reference files and connected apps, will replace Custom GPTs. Virtualization Review reports a general retirement date of December 11, 2026, with timing that varies by workspace; Harvard’s IT team told its community that Custom GPT functionality there ends February 11, 2027. Instructions and knowledge files migrate. Past conversations, the selected model, sharing settings and custom actions don’t, and OpenAI cautions that a migrated plugin “may respond differently.”
To be fair to both companies: Skills and plugins are more capable than what they replace, and Google gave education a long runway. That isn’t the problem.
The hidden cost of platform churn for schools
The problem is that someone else picked the timing. Think about a middle school ELA team that built a shared feedback assistant last year. Under these changes, someone has to find every assistant, rebuild or activate it, re-test it with real student work (with no student names in the test), re-share it with the right people, and update the how-to slides teachers rely on. Multiply that by every department, every instructional coach and every principal who built something useful.
None of that shows up on an invoice, but it is real cost. It lands on your most enthusiastic adopters, the people who did exactly what you asked them to do. And the next change, whether a renamed feature, a new usage limit or a plan restructure, will land the same way, because the district doesn’t own the roadmap.
This is where I keep coming back to a point from my Human Still Required post on the ed-tech backlash: judge every tool by its purpose, cost, evidence and who depends on it. Platform churn is a “who depends on it” problem. The more your staff depends on a tool, the more a vendor’s product decision becomes your workload.
What a district-owned AI platform changes
On a private server, the district’s shared assistants, knowledge bases and permissions live on hardware the district owns, inside the district network. When the platform software gets better, districts get those updates as their server is updated through support, and the district decides how and when to roll new features out to staff and students. You can see what that looks like in our recent post on recent updates to the private server, which covered things like version history with rollback for shared assistants.
It also means you aren’t tied to one company’s model. The server runs open-source models locally, and if the district chooses, it can connect commercial models from Anthropic, OpenAI or Google by API, side by side, under the district’s own rules. If a better model comes out, you add it. Your teachers’ assistants don’t have to be rebuilt because a vendor rebranded a feature.
A quick, honest note on cost, since it always comes up. API use isn’t included in the server price: the district opens its own account with each provider and pays for the tokens it uses. A token is a small chunk of text, roughly three-quarters of an English word, and providers charge per million tokens read and written. Anthropic currently lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens. Hypothetically, a teacher request of 3,000 tokens in and 1,000 out costs about $0.006 + $0.01 = $0.016. A teacher making 200 of those a month would cost about $3.20, compared with a published per-seat price like Google AI Pro for Education at $20 per user per month on an annual plan. I walked through the full math in paying for seats vs. paying for use.
What district leaders can do this month
- Inventory your custom assistants. Ask staff which Gems and Custom GPTs they actually use with colleagues or students.
- Put the dates on the calendar. November 17, 2026 (Gems move into Settings), June 1, 2027 (education Gem access ends no sooner than this), and whatever date your OpenAI workspace shows for Custom GPT retirement.
- Rebuild the critical few early. Pick the five assistants that matter most and rebuild them now, rather than waiting for automated conversion. If you’re moving to Skills, the free Shifting Schools Gemini Skills Library is a good starting point.
- Write down the prompts and files. Keep a district copy of each important assistant’s instructions and reference documents, outside any vendor’s product.
- Add “platform churn” to your AI evaluation rubric. For every tool, ask: if this feature disappeared next year, what would it cost us to rebuild?
Why a private server
A private server isn’t one more chatbot. It’s the district’s own AI platform. Through PrivatEDU AI it’s a one-time $20,000 purchase that includes pre-configured hardware, 12 months of remote support and a year of virtual professional development, with no per-user licensing fees and support for roughly 200 concurrent users. We can run it in a secure cloud, but for security reasons we recommend on-site: prompts and logs stay on district hardware inside the district network, and the district holds the keys.
It’s offered through LeadershipAIplus, a nonprofit, educator-led program of USA-Kansas run by former school leaders. That matters here. When the platform changes, you’re working with educational leaders supporting educational leaders, not a vendor counting seats. Districts from Kansas to Washington to Berlin are already running their own servers this way.
FAQ
When do Gemini Gems stop working for schools?
Google says education editions won’t lose Gem access before June 1, 2027. Gems move into Settings in the Gemini app on November 17, 2026, and remaining Gems are later converted to draft Skills that creators must activate.
What happens to Custom GPTs our teachers built?
OpenAI is replacing them with plugins. Instructions and knowledge files can be migrated, but past conversations, the chosen model, sharing settings and custom actions don’t carry over, so plugins need re-testing and re-sharing.
Does a private server ever change?
Yes, the platform keeps improving. The difference is that updates come to the district’s own server through support, and the district decides how to roll out new features to staff and students.
The bottom line
Gems and Custom GPTs won’t be the last features to change. Renting a platform means living on someone else’s timeline. Owning one means you set it.
So here’s my question: if your most-used AI assistant disappeared next semester, how many hours would it take your staff to rebuild it?
If you’d like to talk through what owning your AI platform would look like in your district, schedule a Systems Review.
Sources
- Google Workspace Admin Help, “About the transition from Gems to skills,” last updated Sept. 30, 2026. knowledge.workspace.google.com
- Jay Bonggolto, “Google outlines timeline to phase out Gemini Gems in favor of Skills,” Android Central, Sept. 30, 2026. androidcentral.com
- David Ramel, “OpenAI to Retire Custom GPTs, Replace Them with Plugins,” Virtualization Review, Sept. 28, 2026. virtualizationreview.com
- Harvard University Information Technology, “OpenAI will retire Custom GPT functionality in ChatGPT,” Sept. 30, 2026. huit.harvard.edu
- Anthropic, “Pricing,” Claude API docs, accessed Oct. 31, 2026. platform.claude.com
- Google Workspace Admin Help, “Google AI Pro for Education,” accessed Oct. 31, 2026. knowledge.workspace.google.com