Google Gemini Skills Replace Gems in Agent Prompt Shift

Google Gemini Skills Replace Gems in Agent Prompt Shift

Google is retiring Gems, its tool for building custom versions of Gemini, and replacing it with "Skills." The new feature is rolling out to Gemini chat users worldwide. With this change, Google joins OpenAI and Anthropic in moving away from personalized chatbots and toward reusable instruction packages that an AI system can call on when needed.

Gems were Google's answer to OpenAI's Custom GPTs. OpenAI is also phasing those out. The format replacing them did not come from Google or OpenAI. Skills started at Anthropic and are now published as an open standard.

What a Skill actually is

A Skill is a detailed prompt written for a specific task. Users can draft one themselves, or have Gemini help write and refine it. The goal is to store instructions you would otherwise retype often, and then call them up on demand.

There are three ways to use them:

  1. Manual invocation. Type "/" followed by the Skill's name, and Gemini applies the stored instructions.
  2. Automatic triggering. When Gemini sees a prompt that matches an existing Skill, it can run that Skill without being asked.
  3. Generation from history. Gemini can look at your previous chats and turn recurring requests into new Skills.

Skills can also be chained. Google's example is a larger writing task that combines one Skill for a personal writing style with another that holds brand guidelines. Each Skill does one job, and stacking them produces the combined result.

Reference material and what comes next

Skills can take reference material along with instructions. Text documents, PDFs and images are supported at launch. Google says several more additions are coming in the next few weeks:

  • Sharing Skills with other people
  • Using files from Google Drive
  • Integration with Gemini Notebook

After those, Skills will roll out to Workspace customers in enterprise, education and nonprofit organizations. Workspace is Google's paid suite of productivity apps for organizations, which includes Gmail, Docs and Drive.

The Gems shutdown timeline

Gems will not disappear for everyone at once. The phase-out has three stages:

  • November: personal accounts
  • March 2027: enterprise and nonprofit Workspace customers
  • June 2027: education customers

Users do not need to rebuild anything by hand. Existing Gems will be converted to Skills automatically.

November also brings the end of another Google project. Opal, the AI mini-app experiment Google launched in summer 2025, will shut down that month as well.

Why the format changed

The change in format matters more than the change in name. Gems and Custom GPTs were built as separate chatbots. You opened a specific one, talked to it, and closed it. Skills work the other way around. The instructions stay in the background and are loaded when the task calls for them, either because the user asks or because the model detects a match.

That fits how agent-style systems are designed to work. An agent working through a multi-step job needs to pick up the right instructions at the right moment. It should not have to send the user to a different persona. Chaining and automatic triggering are both features aimed at that kind of workflow.

The Bigger Picture

Three of the largest AI labs have now landed on roughly the same approach. That suggests the era of the "custom chatbot" is ending, and the GPT store model that OpenAI introduced is giving way to modular instructions. The fact that Google adopted a format that started at a rival and is available as an open standard stands out. If Skills really do work across products, people who write them could reuse their work instead of being locked into one vendor. Whether that portability holds up in practice is still an open question.

Automatic triggering deserves a closer look. When a model decides by itself which stored instructions to apply, users have less direct control over its behavior. This is the same tension that shows up in always-on agent products. It is worth watching how reliably Gemini matches prompts to Skills, and whether sharing Skills creates new ways to misuse AI agents once shared Skills spread through Drive and Workspace. For anyone who depends on Gems now, the next step is simple: check how your Gems come out after the November conversion.