Google starts shifting Gemini customization from Gems to portable Skills

A shift from personas to packages

Gems were designed as customized Gemini experiences. Users could create a tailored assistant for a specific role, task or body of knowledge. Skills are framed differently: They are reusable instruction sets Gemini can apply in the flow of work, either because a user invokes them or because Gemini recognizes that a skill is relevant to the current task.²

That distinction matters. A Gem often behaved like a separately configured assistant. A Skill is closer to an agent configuration artifact: a package of instructions, metadata and supporting files that can sit alongside other skills and be used inside an ongoing workflow.

The central technical marker is SKILL.md, a Markdown-based file that holds the skill’s instructions. Google’s move to that format is intended to make skills more portable across compatible environments, including cases where a skill was created outside Gemini and imported later.¹

For builders, the format should make skills easier to version, inspect and standardize than free-form prompt text buried in a chatbot setting. But portability should not be confused with guaranteed behavioral compatibility. Reports on the migration note that imported or converted skills may still behave differently depending on the Gemini surface, supported tools, file availability and rollout status.⁵

Skills can stack inside Gemini workflows

One of the biggest functional changes is that Skills can be combined. Instead of switching into a single Gem, users may be able to apply multiple skills in the same prompt or conversation, such as a brand-voice skill with a vendor-evaluation skill, or a lesson-planning skill with an institutional policy skill.¹

That stacking model is important for enterprise workflows. It suggests Google wants reusable skills to operate like modular workflow components rather than isolated assistants. A team could maintain separate skills for tone, compliance, formatting and domain expertise, then combine them as needed.

However, Skills are not currently a single universal layer across every Gemini surface. Guidance on manual conversion emphasizes that Gemini app skills and Workspace skills are separate, and that users who need the same guidance in both environments may need to recreate the skill in each place.³

That separation is a key administrative issue. A skill created for the Gemini app should not be assumed to appear automatically in Workspace apps, Workspace Studio flows or other enterprise contexts.³

Migration is not one-to-one

Google plans to phase out Gems gradually and migrate remaining Gemini-app Gems into draft skills later, but admins should not treat that as a parity guarantee. Migration guides emphasize that copying a Gem into a Skill generally means preserving the Gem’s name, description and instructions, then handling knowledge files separately.⁴

Manual migration typically involves opening the Gem, copying its core fields, downloading any knowledge files, creating a new skill, downloading the generated ZIP, placing SKILL.md and the knowledge files in the same correctly named folder, and uploading that folder as the replacement skill.³

That process is more like reconstruction than conversion. It can preserve the material a builder moves, but it does not necessarily preserve every Gem feature, integration or tool behavior.⁴

Migration coverage also depends on file support. Reporting on the transition notes that supported knowledge files may be included in automatic migration, while unsupported file types, linked repositories, internet-dependent scripts or tool-specific behavior may require redesign.⁶

What admins should check first

Workspace administrators should begin with an inventory of Gems that support business-critical workflows. The highest-risk Gems are those that depend on attached knowledge files, Workspace Studio flows, specialized Gem tools, Google Drive or repository-linked material, or instructions that users have refined over time without storing a separate copy.⁵

Admins should also identify Gems used in Workspace Studio flows. Existing flows with an Ask a Gem step may continue for now, but Google’s broader Gems retirement means those flows will need review before Gems stop working for business and enterprise accounts.¹

The safest approach is to prioritize high-value Gems for manual rebuilding and testing, rather than waiting for auto-migration. Builders should keep a local copy of each Gem’s instructions and download any knowledge files before making changes.⁴

Testing should be task-based. A migrated Skill should be checked against the actual prompts, files, outputs and review standards the Gem supported. If a skill produces different results, likely causes include activation differences, missing files, unsupported features, surface-specific behavior or the gradual rollout itself.⁵

Key dates for the transition

The first Workspace milestone is October 5, 2026, when Skills begin rolling out to Rapid Release domains. Scheduled Release domains follow later in October, with Workspace rollout expected to finish by mid-November. Gemini app rollout is scheduled to begin October 13 and also finish by mid-November.¹

Gems will not vanish immediately. Reporting on Google’s schedule says Gems move to the Gemini app Settings panel on November 17, 2026, while users can still create, edit and use them at that stage.³

For Workspace business and enterprise accounts, Gems are expected to become unavailable no sooner than March 1, 2027. For education accounts, the comparable cutoff is no sooner than June 1, 2027.³

Those later dates give organizations time to plan, but the October rollout starts the practical transition. Once users begin creating Skills, admins will need policies for naming, ownership, sharing, file handling, version control and testing.

What workflow builders should do now

For AI workflow builders, the move to SKILL.md creates an opportunity to treat Gemini customizations more like maintainable assets. Teams can standardize instruction templates, keep copies in documentation or repositories, define review processes and separate reusable guidance from one-off prompts.

The format also encourages modular design. Instead of building one large Gem that handles every step in a process, teams may be better served by smaller skills for discrete tasks such as classification, summarization, tone adaptation, data extraction, policy review or output formatting.

But builders should avoid assuming that a Skill imported from another platform, or a Gem recreated as a Skill, will behave identically. Sources tracking the change stress that unsupported internet-dependent scripts, GitHub-file limitations and missing feature parity may affect migrated workflows.⁶

The near-term best practice is to run Gems and Skills in parallel where possible. Keep the original Gem until the Skill has been tested, document any output differences, and update user instructions so employees know whether to invoke a Skill directly, rely on automatic matching or continue using a Gem during the transition.⁵

Why it matters

Google’s Skills rollout is more than a product renaming. It signals a broader shift in enterprise AI customization from bespoke chatbot personas to reusable workflow artifacts that can be packaged, moved, combined and governed.

That should make Gemini workflows more manageable over time. It also creates short-term migration work for admins who need to preserve critical Gems before they disappear.