Google Gems are on the way out. What’s an educator to do?
Honestly, I didn’t even notice the shift I had made in my own work until this week. Somewhere in the last few months, I stopped using Google Gems and Microsoft Copilot Agents inside my productivity tools. I have been more focused building skills. Just last week, I built a skill to edit and format a PowerPoint based on brand guidelines and a template. I tested the skill, confirmed it acted as intended, and then shared the .md file with a colleague. I never even considered building an Agent – it was simply a job to do, inside an app. This might be because I’ve been using Skills inside Claude for a year, and it just made sense to use them once they appeared in Office.
On September 30, Google announced Gems are going away, and they’re replacing Gems will Skills. And they’re going away quickly – personal accounts lose access by next month. Fortunately, education accounts keep access through June 2027, and existing Gems will be converted to Skills automatically.
My first reaction was that this move doesn’t make any sense, because I’m an educator. Then I remembered that I’d made a Skill that could have been an Agent, and that Google and Microsoft are enterprise-level, productivity-first companies. And because I’m also a CEO, it started to kind of make sense.
Here is the shift in two sentences:
A Gem is a place you go. A Skill is something your AI tool can use.
A Skill is a reusable set of instructions the AI can pull in when a task calls for it. Google, Anthropic, and Microsoft are all moving this way. It’s a smart design for work. A marketing team doesn’t need seventeen separate chatbots; it needs one assistant that knows the brand guidelines when it’s writing and the style guide when it’s editing.
But Gems, Agents, and Custom GPTs are great for educators.
When an educator shared a link to the “Constitutional Convention Delegate,” students had a place to go, with a role, a set of sources, and a rule about how the conversation would work.
With a Gem, the teacher decides what the AI does. With Skills, the model increasingly decides which capabilities to load. In fact, Google is saying Gemini can even recognize when a Skill will be relevant. That’s great for productivity. But in a classroom, scaffolds, constraints, and questioning strategies are part of what we design on purpose. Moving those teaching and learning design decisions into a platform makes the experience less human.
Let’s be clear – creating a skill, sharing it with students, and then having them type a slash command into their regular chat window (assuming they can even access the full LLM) can produce the same behavior. It just doesn’t produce the same framing. Plus, Skills currently live in a student’s normal chat history instead of a dedicated space, which could feel confusing. Educators won’t be able to post a URL to their LMS and expect students to access it. It’s going to be more complicated.
There is a big shift here for our thinking as educators. Instead of asking “how do I build a Gem?” we have to start asking, “how do I write down an instructional practice so any AI can use it?”
Skills are built on an open, portable format, which means something you create isn’t trapped in one company’s product. Skills can be stacked, so a “Socratic questioning” skill can work alongside a “primary source analysis” skill.
Where we’re headed next
I expect AI products to divide into two camps. General platforms like Gemini, Claude, Copilot, and ChatGPT will become something like an operating layer, built around tasks, productivity, and efficiency. That’s where the enterprise money is.
Education-native tools have a different purpose and a more human-centered reason to thrive. SchoolAI Spaces, Brisk Boost, and others are built around the idea that the teacher is designing the activity, the student enters, and the teacher can see what happened. Their goal is closer to designing the conditions under which the student does the thinking and the educator has oversight – which was never true in Gems, Agents, or CustomGPTs anyway.
So maybe the big companies own the intelligence layer, but education companies own the design of learning. That could make the education-focused platforms more important, not less.