Google Retires Gemini Gems, Introduces Reusable Skills Feature

Google announced that the Gemini “Gems” feature will be retired starting in November 2026 and will be replaced by a new custom‑instruction system called “skills.” The change applies to all Gemini users, but the migration timeline differs by account type.

Migration schedule by account type

Personal Google accounts will see their Gems automatically converted to skills in November 2026. Workspace business, enterprise and non‑profit accounts will transition in March 2027, while Workspace education accounts will follow in June 2027.

Google states that the automatic conversion will preserve every user’s knowledge files, meaning existing Gems will appear as equivalent skills without data loss. Users who prefer an earlier switch can manually recreate their Gems by downloading the files, creating a skill, and uploading a SKILL.md bundle.

How skills work and current limitations

A skill can be invoked in any Gemini chat by typing a forward slash ("/") followed by the skill name. Google has signaled that an "@" shortcut will be added later, but no release date has been provided.

• Maximum of 100 active skill instances per user at any time
• No hard limit on total created skills
• Skills are currently usable only with Connected Apps
• Skills do not work with Canvas, Deep Research, Create video, Create music, or Guided learning

User experience concerns

TechCrunch criticizes the shift, arguing that a slash‑command interface feels more like a developer tool than a consumer‑friendly feature. The outlet also notes that Google’s history of rebranding can create confusion among existing users.

The documentation from Google Support confirms the three‑step manual migration process but does not specify when the "@" trigger will become operational or which default Gemini tools will gain skill compatibility in future updates.

There is also uncertainty around the 100‑skill active limit. Google has not indicated whether the cap will be adjusted after the migration period ends, leaving power users to monitor their active skill count closely.

Overall, the transition represents a strategic move to make Gemini interactions more modular and developer‑oriented, while still promising seamless data preservation for everyday users.

The migration from Gems to the reusable skills system hinges on the premise that each user’s existing knowledge files will be faithfully re‑encoded as skill bundles. This claim necessitates a verification step: after the automatic conversion, users should audit a representative sample of their former Gems to confirm that the functional behavior, context retention, and any embedded prompts remain intact. Any discrepancy would indicate a flaw in the conversion algorithm, prompting a rollback or manual remediation. By emphasizing data preservation, Google seeks to mitigate the risk of user churn that could arise from perceived loss of invested content, yet the onus of validation now rests with the end‑user, especially in environments where critical workflows depend on precise prompt configurations.

The imposed ceiling of one hundred active skill instances per user introduces a hard operational limit that could constrain power users and organizational deployments. While there is no restriction on the total number of skills that can be created, only a subset may be active simultaneously, meaning that users must prioritize which skills are loaded into their current session. This limitation may compel administrators to implement governance policies—such as periodic reviews of skill usage metrics—to deactivate dormant skills and free capacity. The uncertainty surrounding any future adjustment of this cap adds a layer of strategic planning, as organizations must anticipate potential scaling needs and design fallback mechanisms before the limit becomes a bottleneck.

Current limitations also stem from the fact that skills are presently usable only with Connected Apps, excluding several high‑visibility Gemini functionalities like Canvas, Deep Research, video creation, music generation, and guided learning. This restriction narrows the practical applicability of skills to a subset of use cases, effectively relegating the feature to a niche rather than a universal enhancement. Users seeking to leverage skills within the excluded modules will encounter a fragmented experience, potentially leading to workarounds that undermine the intended modularity. Continuous monitoring of feature rollouts will be essential to gauge when, if ever, broader integration occurs.

The user‑experience shift toward a slash‑command invocation model marks a departure from the more intuitive, conversational approach that characterized Gems. Although a future "@" shortcut is promised, its timeline remains undefined, leaving the current interface feeling more akin to a developer console than a consumer‑friendly tool. This perception may affect adoption rates, particularly among non‑technical users who might find the syntax barrier intimidating. Training programs therefore need to incorporate hands‑on exercises that demystify the command syntax, reinforce best practices for naming conventions, and illustrate how to efficiently toggle active skills within the prescribed limit.

From an organizational standpoint, the staggered migration schedule creates a bifurcated planning horizon: personal accounts transition in late 2026, while Workspace business, enterprise, non‑profit, and education accounts follow in 2027. IT departments must align their rollout calendars with these dates, ensuring that critical Gems are identified, tested, and re‑created as skills well before the cut‑over. Additionally, the manual migration pathway—downloading Gem files, constructing a SKILL.md bundle, and uploading it—requires a clear procedural checklist to avoid data loss or configuration errors. Auditing tools that compare pre‑ and post‑migration skill behavior can serve as a verification layer to certify successful transitions.

Overall, the strategic pivot to a modular, developer‑oriented skill architecture aims to future‑proof Gemini interactions by enabling reusable, shareable components. However, the practical consequences include a need for rigorous validation of data fidelity, proactive management of active skill quotas, and comprehensive user education to bridge the usability gap introduced by the slash‑command paradigm. Organizations that invest in systematic testing, governance frameworks, and targeted training will be better positioned to harness the promised flexibility while mitigating the operational risks inherent in this transition.

For English‑speaking organisations, the concrete impact is twofold: IT teams must plan the March 2027 (or June 2027 for education) migration to ensure business‑critical Gems are available as skills, and end‑users will need to adapt to the slash‑command workflow. Training programs should focus on creating and managing skill bundles, and monitoring the 100‑active‑skill limit to avoid service interruptions.