Google is shutting down Gems, the feature that let Gemini users build their own task-specific AI assistants, and replacing it with something called 'skills.' If your business built a custom Gem to draft customer emails or summarize call notes in your voice, that setup is on its way out.
Gems launched as Google's answer to custom GPTs: a way to configure a version of Gemini with specific instructions, tone, and reference documents, then reuse it like a saved template. A marketing team might build a Gem trained on brand guidelines. A solo consultant might build one that formats invoices a certain way. It was, in effect, a way to make a general chatbot behave like a specialist without any coding.
Skills work differently. Instead of spinning up a separate, named assistant for each task, the idea is that a single Gemini agent picks up and drops modular capabilities as needed, the way a person might grab a different tool from a drawer without becoming a different person. Google has not published a full technical breakdown, but the shift points toward one assistant with pluggable functions rather than a stable of user-built bots.
The timing lines up with moves elsewhere in the industry. Meta recently pushed its Muse and Instinct assistants as all-purpose agents rather than a collection of narrow tools, and other major AI vendors have spent the past year consolidating their product lines around one flagship assistant instead of a family of specialized variants. Building and maintaining dozens of separate custom-bot features is expensive; funneling users toward one adaptable agent is cheaper to support and easier to market.
Why it matters
This fits a pattern that's shown up repeatedly since custom GPTs and their imitators launched in 2023: companies roll out a customization feature to drive engagement and developer interest, then quietly narrow or retire it once a cleaner, more centralized product direction emerges. Custom-bot marketplaces have generally struggled to generate the usage numbers vendors hoped for, and maintaining them alongside a core product creates duplicate engineering work. Consolidating into one configurable assistant is the more common outcome, not the exception.
What this means for small businesses
If you invested time building Gems for repeatable tasks, treat this as a migration project, not a feature update. Document exactly what each Gem does, what instructions and files you fed it, and what output format it was set to produce, because there's no guarantee that configuration transfers automatically to the new skills system.
There's also a broader lesson here about where to put your workflow eggs. Custom configurations built inside any single vendor's tool, whether that's a Gem, a custom GPT, or a workflow automation, are subject to that vendor's product roadmap, not yours. Businesses that keep a written record of their prompts and setups outside the platform tend to recover faster when a feature gets restructured.
On the upside, if skills genuinely let one assistant handle more without you switching between saved bots, day-to-day use could get simpler. Fewer clicks to find the right assistant is a real, if modest, convenience gain.
What to watch
Watch for Google's migration deadline and whether it offers an export or auto-conversion tool for existing Gems. Also watch whether skills become an open system that outside developers can build and sell into, similar to app-store models other AI vendors have tried, since that would affect pricing and the range of tools available to small businesses down the line.
The bottom line
If your business relies on a custom Gem, back up its instructions and test files now, and check Gemini's release notes before any stated cutoff date so you're not caught rebuilding a workflow under time pressure.