Gems are turning into Skills, and they’re moving into your chat
Starting November 17, Google is initiating a significant transition for personal account holders: Gems, the company’s standalone custom AI bots, are being phased out in favor of a new format called Skills. While Gems previously required users to open a dedicated, walled-off interface to interact with a specific persona or workflow, Skills represent a shift toward integrated, modular AI.
The functional difference is notable. Instead of navigating to a sidebar to select a bot, users will access their custom instructions directly within the main Gemini chat interface by typing a forward slash in the prompt bar. A significant enhancement here is the capability for “stacking,” where multiple Skills can be applied to a single conversation. For example, a user could simultaneously apply a Skill that dictates a specific brand voice alongside another that enforces a particular data formatting style. This eliminates the need to cycle outputs between different windows.
However, there are trade-offs to consider during this transition. Many advanced features that were previously toggleable within Gems—such as Deep Research, Canvas, and direct video or music generation—are not yet compatible with the new Skills framework. Furthermore, the ability to share a Skill or attach files from Google Drive is currently absent, though Google has indicated these features are in development. Users who rely on these specific tools are advised to audit their workflows before the November 17 deadline.
Scheduled Actions are prompts with an alarm clock
For users who need consistent updates without real-time interaction, Scheduled Actions serve as a time-based automation tool. This feature allows a user to write a prompt and define a frequency for its execution. Gemini then prepares the output in the background, delivering the result via push notification on mobile devices or as an unread chat item on the web.
Scheduled Actions are currently available to free accounts, provided that the user has “Keep Activity” enabled. The system functions by generating the response in the hour preceding the scheduled delivery time. Consequently, this tool is most effective for static information retrieval, such as a morning calendar briefing or a daily to-do list summary, rather than volatile data that shifts significantly minute-to-minute.
A critical limitation involves location-based prompts. Because the system utilizes the location data active at the moment the action was configured, it does not dynamically track a user’s movement. If you set a daily weather report for a specific city, you will continue to receive data for that location regardless of where you travel, unless you manually update the action.
Spark is the agent that keeps working after you log off
While the naming conventions at Google remain fluid, the background agent formerly known as Spark is currently being rebranded as “Task” within the Gemini ecosystem. This tool acts as an always-on, autonomous assistant running on Google Cloud virtual machines. Unlike the aforementioned tools, this agent operates independently of the user’s active session, meaning it continues to execute tasks even after a laptop is closed or a phone is locked.
The utility of this agent lies in its ability to handle complex, high-level objectives that involve multiple steps across different applications. When given a goal—such as comprehensive competitive research or monitoring web-based events—the agent performs the legwork and periodically checkpoints with the user. To prevent unauthorized actions, the system is designed to pause and request manual confirmation before executing sensitive steps, such as making a purchase, submitting a web form, or accessing password-protected areas.
Given the autonomous nature of this agent, Google recommends that users remain vigilant regarding the outputs. The system currently supports up to 15 concurrent tasks and requires an active Google AI Pro or Ultra subscription to function.
Why scheduling a Spark task isn’t a Scheduled Action
The terminology can be misleading, as both Scheduled Actions and the “Task” agent can operate on a timer. However, the underlying mechanics differ significantly. A Scheduled Action is essentially a static prompt triggered by a clock, whereas a Task (formerly Spark) is event-driven and capable of multi-step autonomous execution.
For instance, while a Scheduled Action might check your email at 8:00 AM every day, an autonomous Task can be configured to monitor for specific flight price drops and alert you only when a fare meets your defined criteria. Furthermore, the Task agent is capable of performing actions across various applications, such as unsubscribing from newsletters or updating spreadsheets, whereas Scheduled Actions are primarily focused on information delivery.
Back up your prompts while you can
As Google continues to refine its AI suite, the longevity of any specific interface or naming convention remains uncertain. Skills offer a flexible, free way to manage custom instructions, while Scheduled Actions provide a robust, no-cost method for receiving daily information. The higher-tier “Task” agent remains the most powerful choice for users requiring cross-application, event-driven automation.
To ensure your workflows remain intact despite future platform shifts, it is recommended that you perform a manual backup of your existing Gems before November 17. Copying your system instructions into a Markdown file and downloading any associated knowledge files will safeguard your hard work. By keeping these prompts in your own repository, you insulate your processes against future UI updates or service sunsets.
Disclaimer: This content is auto-generated for informational purposes only.
Source: Read Original News
