Google Deprecates Gemini Gems in Favor of New Skills Feature

Google Reshapes Gemini's Feature Lineup

Google is making a notable change to its Gemini AI assistant: it's sunsetting Gems — the feature that let users build task-specific AI agents — and replacing it with a new mechanism called "Skills." This shift reflects a deeper change in how AI assistant products are being designed, moving away from manually assembling multiple custom agents toward a more unified, general-purpose approach.

Gems were originally designed to let users create specialized AI personas or assistants for specific tasks. For example, you could configure separate Gems for writing, coding, fitness planning, and so on — each with its own preset instructions and context. In essence, the model fragmented complex needs into a collection of discrete, custom-built assistants.

Why the Switch to Skills

The backdrop to this decision is the rapid rise of all-in-one AI agents in the market. As the original report notes, the popularity of products like Meta's Muse and Instinct — capable, general-purpose agents — has prompted Google to move away from the "one agent per task" paradigm.

The underlying logic isn't hard to follow. For everyday users, creating and managing multiple Gems for different task types carries a steep learning curve and a high barrier to entry. The more appealing experience is a smart, general-purpose assistant that automatically draws on the right capabilities based on context — without requiring users to constantly switch between custom agents.

From "Manual Assembly" to "Capability Invocation"

The name "Skills" itself reveals a new design philosophy. Rather than treating "an entire assistant" as the unit of management like Gems did, Skills are more like mountable capability modules attached to a unified Gemini assistant. Users no longer need to maintain a fleet of independent agents; instead, they rely on a single primary assistant that triggers different skills on demand.

This architecture delivers a more seamless experience and better aligns with the industry's vision of a "general-purpose agent" — a single entry point, multiple capabilities, and automatic context-driven responses.

Meta's Muse and Instinct are two distinct general-purpose agent products under Meta AI. Muse focuses on creative and content generation use cases, while Instinct leans toward task execution and personal assistant functions. What they share is that neither requires users to pre-configure a specific persona — the model itself determines the appropriate response mode from conversational context. This stands in sharp contrast to Gems' "build first, chat later" model, and represents the mainstream direction major tech companies are taking with agent products: encapsulating capabilities within the model itself, rather than exposing them for users to assemble manually.

From an architectural standpoint, the Skills mechanism is closely aligned with the widely adopted "Tool Use / Function Calling" design pattern. In this architecture, the large language model acts as a central orchestrator, dynamically selecting and executing specific functional modules — such as web search, code execution, or calendar management — based on user intent. Users don't need to know what's happening under the hood; the model decides "which skill to use and when." This is fundamentally different from Gems, which required users to actively choose and switch to a specific assistant. From the user's perspective, it significantly reduces cognitive load and comes much closer to the intuitive expectation of what a "truly intelligent assistant" should feel like.

Industry Signal: The Consolidation of AI Assistants

Google's adjustment can be seen as a microcosm of the broader consolidation happening across the AI assistant space. Early AI products tended to offer a large number of customizable components, letting users stitch together their own workflows. But as underlying model capabilities have grown stronger, product teams feel more confident hiding that complexity on the backend and delivering a simpler, more capable unified experience to users.

The fact that Meta's Muse and Instinct are called out as driving this trend suggests that major tech players have reached a kind of consensus on what AI agent products should look like: rather than turning users into "agent builders," make the assistant itself powerful and flexible enough to handle the job.

For users already relying on Gems, this change means adapting to a new way of working. But in the long run, if Skills can truly deliver seamless capability orchestration, the reduction in overall friction will be a tangible benefit.

Summary

Google replacing Gems with Skills is a clear signal that product thinking is shifting from "fragmented customization" to "unified intelligence." As all-in-one AI agents continue to gain traction, this kind of adjustment may just be the beginning. For those watching how AI products evolve, the key question going forward is whether Skills can deliver on its promise of being smarter and more effortless in practice.

(Note: This article is based on a single-source report. For specific feature details and release timelines, refer to official Google announcements.)