Google Kills Gemini Gems, Replaces Them With 'Skills'

Google is retiring Gems, the custom chatbot personas in Gemini, and replacing them with "Skills" — detailed, reusable prompts that users can invoke by typing "/" in Gemini chat, and that Gemini can also run automatically on its own. Gems will begin phasing out in November, and Google says existing Gems will migrate automatically to the new format. The Skills format is based on an open standard that originated with Anthropic.

That last detail is the story. Google did not invent a proprietary format for reusable prompts. It adopted an open standard from Anthropic, its closest competitor in the frontier-model race. And in doing so, it joins OpenAI and Anthropic in an industry-wide shift away from chatbot personas and toward what are best described as agent-ready prompt formats.

What is actually changing in Gemini chat

For users, the change is straightforward. Where a Gem was a persistent, persona-style customization of Gemini — a saved configuration that shaped how the assistant behaved across conversations — a Skill is a structured, reusable prompt. The user invokes it deliberately by typing "/" followed by the Skill, or Gemini triggers it automatically when the system decides the Skill applies to the task at hand.

That second mechanism is the meaningful one. A prompt format that a model can invoke on its own, without the user explicitly calling it, is not a saved preference. It is closer to a callable instruction — a unit of behavior the assistant can select and execute as part of a longer chain of actions. This is the mechanical difference between a chatbot customization and an agent-ready component, and it is why the shift is happening at three major labs at once.

Google has committed to a migration path rather than a hard cutoff. Gems will be phased out starting in November, and existing Gems will move to the Skills format automatically, so users should not lose their saved configurations in the transition.

Why an open standard matters here

The most consequential sentence in Google's announcement is easy to skim past: the Skills format is based on an open standard from Anthropic. The company that makes Claude defined a specification for reusable, structured prompts, published it openly, and Google — the company that spent years building its own parallel everything in AI — adopted it rather than proposing a rival format.

That decision carries weight beyond Gemini chat. Prompt formats only become infrastructure if they are portable. A user who invests time writing detailed Skills, or a team that builds a library of them for repeatable workflows, can only treat that library as an asset if the format survives contact with other models and other providers. An open standard creates that possibility. A proprietary one does not.

For Google specifically, adopting an external standard marks a pragmatic turn. The company has the resources to define its own format and push it. Choosing Anthropic's open standard instead signals that Google judged interoperability in the prompt layer to be worth more than control of it — at least at this stage, while the conventions for agentic AI are still settling.

The industry pattern: OpenAI and Anthropic are moving the same direction

Google is not out on a limb here. The shift has been underway across the major providers, and the framing is consistent: reusable prompt structures that models can invoke and chain are becoming the default way to package repeatable AI behavior. OpenAI and Anthropic have both moved toward formats of this kind, and Google's adoption of an Anthropic-originated open standard now makes the direction explicit at all three companies that define the consumer AI assistant market.

The stakes are easy to state. If Skills-style formats become the standard unit of customization and automation, they become the place where user effort accumulates — the libraries people build, the workflows companies encode, the small economy of shared prompt components that already exists informally around these tools. Whoever controls or shapes that format controls a layer of the stack that sits between the model and the user's intent. An open standard governed collectively keeps that layer from being locked to any single provider. Google signing on to Anthropic's standard is a vote for the open outcome, at least for now.

What it means for Gemini users

In the near term, Gemini users will see Skills replace Gems in chat, with the "/" invocation as the primary manual entry point and automatic triggering as the system-side behavior. November is the month to watch: that is when the phase-out begins, and when users will see how the automatic migration handles their existing Gems.

The longer-term signal matters more than the interface change. Every major assistant vendor is converging on prompt formats designed for autonomous invocation — formats built for agents, not personas. Google dropping Gems for Skills is one more confirmation that the industry has decided the future of consumer AI customization is not a saved personality, but a reusable instruction the model can pick up and run on its own.