AI Professional Development

Gems Are Becoming Skills: Why AI Training Can’t Be One-and-Done

Google is retiring Gemini Gems and OpenAI is retiring Custom GPTs. For schools, it’s proof that AI training built on features expires, and that ongoing, embedded professional learning is what lasts.

By Jeff Utecht · · 9 min read

Instructional coach and two teachers working together at a laptop after school, part of ongoing AI professional development

If your teachers spent last year building Gemini Gems for lesson planning, feedback or differentiation, here’s the news: those Gems are on their way out. It’s the clearest case I’ve seen yet for ongoing AI professional development.

Google’s admin help page on the transition from Gems to Skills, updated this week, lays out the timeline. Skills began rolling out in Workspace on October 5 and in the Gemini app on October 13. On November 17, Gems move into Settings. And for Education editions, no sooner than June 1, 2027, users can no longer create, edit or use Gems at all, and they disappear from Google Classroom and from LMS platforms connected through Gemini LTI.

Meanwhile, on the other side of the street, OpenAI is retiring Custom GPTs and moving to plugins, with a retirement date of December 11, 2026, according to Campus Technology.

So here’s my question for every school leader reading this: what did your AI professional development actually teach your people? If the answer is “how to build a Gem,” you just learned something important. The features expire. The thinking shouldn’t.

Key takeaways

What’s actually changing (and why it matters more than it sounds)

On paper, this is a tidy migration. Google says remaining Gems in the Gemini app will automatically migrate to draft Skills, keeping their names, instructions and knowledge files. That’s genuinely helpful.

But read the fine print the way a principal would. Creators have to review and enable each draft before it works. Skills don’t currently sync between the Gemini app and Workspace, so anything used in Workspace has to be rebuilt. And teachers who used Gems inside Classroom are being pointed toward Gemini Notebook and an upcoming study helper instead.

The OpenAI side looks similar. Per Campus Technology’s reporting, instructions become a skill and knowledge files become reference files, but prior sharing settings don’t carry over, and existing GPT users don’t automatically get access to the new plugin. OpenAI itself warns that “a migrated plugin may respond differently.”

Now picture the teacher who built a feedback assistant and shared it with her grade-level team. Picture the coach who built six of them for the whole department. Every one of those has to be found, checked, re-tested and re-shared. Multiply that across a school. Then across a district.

That’s not a tech problem. That’s a capacity problem.

Ongoing AI professional development is the only kind that holds up

I’ll admit something: I don’t think the Gems-to-Skills change is bad. Skills look more flexible, and the open format they use is a step forward. The issue isn’t that platforms change. It’s that a lot of school AI training was built as if they wouldn’t.

And use is growing fast. A new Gallup and Walton Family Foundation survey found that 59% of K-12 educators used AI tools during the 2025–26 school year, and 64% of those users said it freed up more time for the parts of teaching that matter most. That’s real value. It’s also a lot of workflows now sitting on top of features that can be renamed, moved or retired with a few months’ notice.

School leaders already know what the research says about this. The Learning Policy Institute’s review of 35 methodologically rigorous studies identified seven features of effective professional development. Two of them should be on every AI plan: it “provides coaching and expert support” and it “is of sustained duration.” Teachers need time to try something, use it with real students, reflect and adjust. A single workshop can start that cycle. It can’t finish it.

I’ve lived through this before. I’ve been doing long-term partnerships with schools since 2009, starting with 1:1 laptop rollouts. What I learned then is what I’m seeing now: the devices didn’t change learning. Sustained support for teachers did. The schools that treated laptops as a one-time purchase plus a one-time training got expensive notebooks. The schools that treated it as a multi-year learning journey got changed classrooms.

Where schools usually go wrong

Here are the patterns I see most often when a platform shift like this hits:

What school leaders can do this month

  1. Run a quick inventory. Send a two-question form: “What Gems or Custom GPTs do you use regularly?” and “Do students use any of them?” You’ll learn more about your real AI adoption from this than from any survey.
  2. Share the dates, plainly. Google’s own admin guidance recommends telling users the key dates for their edition. Put November 17 and the June 2027 Education date in a staff bulletin, with one sentence on what it means.
  3. Prioritize the student-facing tools first. Anything students use through Classroom or your LMS needs a plan before anything else.
  4. Turn rebuilding into learning. Pair a teacher who built a Gem with one who hasn’t, and have them recreate it as a Skill together. Free starting points like the Gemini Skills Library can help teams see what a well-built Skill looks like.
  5. Re-test, don’t assume. Ask creators to run the same three prompts through the old and new versions and compare. That’s an AI literacy lesson hiding in a migration task.
  6. Ask what your PD taught. At your next leadership meeting, ask: “If every tool we trained on disappeared tomorrow, what would our people still know how to do?” Our AI for teachers guide is organized around workflows rather than features for exactly this reason.

Why this is partnership work

Every school will get through this migration. The question is whether you come out the other side with stronger people, or just with rebuilt tools that will need rebuilding again in eighteen months.

That’s where a long-term, embedded partner earns their place. Across a year, that might look like a leadership team session in the fall on governance and priorities, coaching cycles with instructional coaches so they can support colleagues, hands-on work with teachers built around judgment rather than features, a parent night that explains what’s changing and why, and a short board briefing so policy keeps up with practice. Face-to-face, virtual or a combination, depending on the outcomes you’re after. Every contract is unique, which is why it’s a partnership.

Continuity is the point. When Google or OpenAI moves the furniture next spring, you want someone who already knows your staff, your systems and where each team is, not a new presenter starting from zero. I bring 25 years in education and an international lens from working with US districts and international schools, and no vendor agenda. The goal is to build your internal capacity, not replace it. My work with the Alaska Staff Development Network is a good example of a partnership growing over time: a five-part series, then a second cohort, then advanced sessions, as described here.

One returning workshop participant put it better than I can:

“I think this time around I have a better focus and understanding of how to use it. Last time it was really cool just learning about all the cool things, now I know how to use it better and am more confident.”

That’s what the second, third and fourth sessions do. Tricia Friedman and I talked about this shift from one-off workshops to sustained learning on the Shifting Schools episode AI in Schools Is Changing: What Leaders Need Now.

FAQ

When do Gemini Gems stop working for schools?

According to Google’s admin help page, for Education editions no sooner than June 1, 2027, users can no longer create, edit or use Gems, and Gems are removed from Google Classroom and Gemini LTI-connected platforms. Gems move to Settings in the Gemini app on November 17, 2026, but still work there.

Do teachers need to rebuild their Custom GPTs?

OpenAI is moving Custom GPTs to plugins, with a retirement date of December 11, 2026, according to Campus Technology. Migration tools exist, but sharing settings don’t carry over and outputs may differ, so teachers should re-test and re-share anything students or colleagues depend on.

What does effective AI professional development for teachers look like?

Research on effective PD points to learning that is sustained, collaborative and supported by coaching. For AI, that means building transferable skills such as writing clear instructions, evaluating outputs and deciding when AI is the wrong tool, rather than training on a single feature.

The skill that lasts

Gems were never the point. Neither are Skills, or plugins, or whatever arrives next fall. The point is a staff that can adapt with confidence when the tools shift under their feet.

So here’s my question for you: if your AI tools changed overnight, would your people be ready, or would they be starting over?

If you’d like to build that kind of capacity across your school community, book a 30-minute call with me. Every partnership is tailored, so reach out and we’ll talk through pricing and what a partnership could look like for your school. You can also email me at jeff@jeffutecht.com.

Sources