Google Retires Gemini's Gems, Betting on 'Skills'
Google is shutting down Gemini's Gems, the user-built task-specific agent feature, and moving to 'skills' instead — a pivot that lands as Meta's Muse and Instinct agents gain ground.
Google is discontinuing Gems, the Gemini feature that let users build task-specific AI agents, and replacing it with a system it calls 'skills,' as all-in-one agents such as Meta's Muse and Instinct gain traction.
Google is ending Gems, the Gemini feature that let people assemble their own narrow, task-specific AI agents, and folding that capability into something it calls "skills." The change, TechCrunch reported Monday, arrives as all-in-one assistants — Meta's Muse and Instinct among them — pull attention away from the do-it-yourself agent-builder model that Gems represented.
Shares of Alphabet (NASDAQ: GOOGL) were quoted at $342.74, down 0.34% on the day, as of the last trade at 17:40 GMT on Monday, against a previous close of $343.92 and an intraday range of $339.56 to $343.59. The move was smaller than the broad tech tape: the Nasdaq 100 tracker (QQQ) was at $738.20, off 0.85%, while the S&P 500 proxy (SPY) traded at $767.25, down 0.53%. Meta Platforms (NASDAQ: META), whose agents are the competitive reference point here, was the notable loser of the session at $723.29, down 3.77% from a $751.66 close, with a day range of $717.25 to $750.58.
What Gems was, and why it is being wound down
Gems were, in effect, saved instruction sets. A user could configure a version of Gemini for one repeated job — a specific writing voice, a recurring research routine, a coding helper tuned to house style — and return to it. It was the consumer-facing cousin of the custom-assistant idea that spread across the chatbot industry: give people a lightweight way to package a prompt as a product.
The problem with that design is that it puts the configuration burden on the user. Every Gem is only as good as the instructions someone thought to write, and most people write one and never revisit it. "Skills," by contrast, implies capability the assistant carries by default and invokes when the task calls for it, rather than a separate object the user has to remember to open. That is a different bet about where the value sits: not in the user's ability to specify an agent, but in the model's ability to recognise what is being asked and reach for the right tool.
The all-in-one agent is winning the framing war
The competitive backdrop matters more than the feature mechanics. Muse and Instinct, Meta's agents, are built around the premise that one assistant should handle the whole job — the kind of general-purpose agent that removes the need for a library of specialised ones. If that model is what users gravitate toward, a directory of hand-built Gems starts to look like a filing cabinet: organised, and rarely opened.
Google's response is consistent with how platform companies typically resolve this. When a user-configurable feature and a native capability compete for the same job, the native capability usually wins, because it works for everybody who never configured anything. Consolidating Gems into skills reduces surface area in the Gemini interface and moves the differentiation back to the model.
There is a cost. Users who invested time building Gems lose a workflow they curated themselves, and the migration path will determine how much goodwill that burns. Custom-assistant features tend to attract the most engaged tier of users — the people who evangelise a product internally at work. Those are exactly the users a company least wants to inconvenience.
What this changes for developers and enterprise buyers
For anyone who built a business process around Gems, the practical question is whether skills are addressable in the same way: can a team define one, share it, and rely on it behaving consistently? A user-created Gem is a portable artifact. A skill the assistant selects on its own is harder to audit, and auditability is what enterprise buyers ask about first.
A skill the assistant selects on its own is harder to audit, and auditability is what enterprise buyers ask about first.
That tension — capability versus controllability — is now the central argument in agent design across the sector. Vendors are pushing toward agents that decide more on their own, while the people signing procurement contracts want to know exactly what the software will and will not do. A shift from explicit, user-defined agents to implicit, model-selected skills moves the dial toward capability. Google will have to answer the control question separately.
Reading the market reaction
Nothing in Monday's price action can be pinned to a Gemini feature retirement. Alphabet's 0.34% decline was milder than the Nasdaq 100's 0.85% drop, which is closer to a stock holding up in a soft tech session than a stock being marked down on product news. Meta's 3.77% fall, on a day when its agents are cited as the reason a rival is changing course, is a useful reminder that agent momentum and share prices are not the same signal.
The Dow 30 tracker (DIA) at $515.31, off 0.42%, sat between the two broad benchmarks, which is the usual pattern when the selling is concentrated in large-cap technology rather than spread across the whole market.
What to watch from here
- Migration terms. Whether existing Gems are converted, archived, or simply switched off will tell you how much Google values that power-user cohort.
- Whether skills are shareable. If teams can publish and version a skill, this is a repackaging with an upgrade attached. If they cannot, it is a simplification that costs enterprise flexibility.
- Meta's cadence. Muse and Instinct are the stated reason the all-in-one model is ascendant. How quickly they add capability sets the pace Gemini's skills have to match.
- Third-party reach. Agents become sticky when they can act inside other companies' systems. The winner of the assistant race is likely to be whichever agent gets the widest integration footprint, not whichever has the longest feature list.
Consumer AI has now been through enough of these cycles to recognise the pattern: a feature ships, a segment of users builds real habits on it, and the platform absorbs it when the underlying model gets good enough to do the job without instruction. Gems is the latest to make that trip. The open question is whether skills feel like more capability or just fewer buttons.
Key facts
- GOOGL price: $342.74, -0.34%, as of 17:40 GMT Sept. 28, 2026
- META price: $723.29, -3.77% on the day
- Feature retired: Gemini Gems, which built task-specific agents
- Replacement: A capability system Google calls 'skills'
Frequently asked questions
What were Gemini Gems?
Gems were a Gemini feature that let users create their own task-specific AI agents by saving a set of custom instructions. A person could configure one version of Gemini for a repeated job — a research routine, a writing voice, a coding helper — and return to it later. Google is now discontinuing the feature in favor of 'skills.'
Why is Google replacing Gems with skills?
The shift follows the rise of all-in-one AI agents, including Meta's Muse and Instinct, which handle a broad range of tasks without the user having to configure a separate agent for each one. Skills imply capabilities the assistant carries natively and invokes as needed, rather than objects a user must build and remember to open.
How did Alphabet shares trade on the day of the news?
Alphabet (NASDAQ: GOOGL) traded at $342.74 as of the last trade at 17:40 GMT on Sept. 28, 2026, down 0.34% from a previous close of $343.92, with an intraday range of $339.56 to $343.59. That was a smaller decline than the Nasdaq 100 tracker, which fell 0.85% the same session.
What are Meta's Muse and Instinct?
Muse and Instinct are Meta's AI agents, cited as examples of all-in-one assistants gaining traction — agents designed to handle an entire task rather than being narrowly specialised. Their momentum is the competitive backdrop against which Google is retiring Gemini's user-built Gems feature.
What happens to Gems that users already built?
The migration terms have not been detailed in the available reporting. Whether existing Gems are converted into skills, archived, or switched off outright is the key practical question for users and teams that built workflows around the feature, and it will determine how disruptive the change proves in practice.
Why does this matter to enterprise customers?
A user-created Gem is an explicit, auditable artifact that a team can define and share. A skill the assistant selects on its own is harder to audit, and auditability is typically the first question enterprise procurement asks. The shift moves Gemini toward greater capability but raises separate questions about control.
