Skills Leave the Chat Box: Google Workspace Makes Custom AI a Native App Interface

For most of the past two years, customizing an assistant meant opening a chat product and teaching it there: a Gem, a custom GPT, a…

skills-leave-chat-box-hero

For most of the past two years, customizing an assistant meant opening a chat product and teaching it there: a Gem, a custom GPT, a saved instruction that lived in the chatbot’s own house. On November 10, 2026, Google took a more consequential step. Skills in Google Workspace — reusable, custom instructions that tell Gemini to apply a team’s rules, templates, and reference files — are now rolling out across Google Workspace Studio and the Workspace apps themselves.

Google describes this wave in the language of general availability. Rapid Release domains began a gradual rollout, lasting up to 15 days for feature visibility, on November 9. Scheduled Release domains follow from November 30. Admins, the company adds, can now monitor and audit skill activity in the Admin console using Workspace Studio log events — a capability it ties specifically to general availability.

The story that matters here is not orchestration, and it is not the assistant as a work operating system. It is the customization interface moving to where the work already lives: Gmail, Docs, Slides, Drive, and Chat, with an @ mention, an automatic trigger, or a Studio flow, instead of a separate chat box the user has to remember to visit.

Custom AI, in the app chrome

Google’s definition is plain. A skill is a set of customizable, reusable instructions you give Gemini to accomplish a task — a brand-voice guide, a template-filling routine, a review checklist grounded in files stored in Drive. Once it exists, it is meant to travel with you across Gemini in Workspace apps rather than staying trapped in a single conversation.

There are several ways to create one. Teams can build a skill from scratch or from a turnkey template in Google Workspace Studio, including by editing Markdown-native SKILL.md files directly in Studio, and can test how it drafts an email, a document, or a presentation before turning it on. They can also draft and collaborate on a skill in Google Docs, then send it to Studio. Or they can stay put: Google says you can prompt Gemini directly in Gmail, Docs, Slides, Drive, and Chat to create, refine, and test a skill without leaving the app you are already in. The Help Center’s example is a Gmail-side-panel prompt that creates a skill to scan incoming customer mail and draft a reply from a standard refund policy.

The supported surfaces are documented with unusual precision, and the gaps are as important as the inclusions.

  • Creating, editing, testing, and sharing skills is available on the web.
  • Invoking skills is supported on the web in Workspace Studio, Gmail, Docs, Slides, Drive, and Chat, and in the Google Chat mobile app on Android and iOS.
  • Skills in Workspace are not yet available in Google Sheets or Google Meet.
  • Skills currently support English prompts and instructions only.
  • Availability is limited to users age 18 and older, on Business Starter, Standard, and Plus; Enterprise Starter, Standard, and Plus; Google AI Pro for Education; and the AI Expanded Access add-on.

Skills in Workspace Studio and Gemini in Workspace are on by default for those eligible editions, and can be turned off at the domain, organizational unit, or group level. External sharing follows the organization’s Studio sharing controls. That is a product-surface decision as much as an admin one: custom AI is being treated like any other Workspace feature that appears in the toolbar unless someone disables it.

@mention, automatic use, and a hook into Studio

Once a skill is turned on, Google offers two ways to use it in the apps, plus a link into automation.

The first is implicit. Gemini in Gmail, Docs, Slides, Drive, and Chat can automatically recognize when a turned-on skill is relevant and apply it to a prompt. The second is explicit: a user can @ mention one or more skills at any time, including combining several skills in a single request. The Help Center describes the mechanics without romance. In the “Ask Gemini with skills” panel, type @, pick the skill, then enter the prompt — or just describe the goal in natural language and let Gemini choose.

The third path is Workspace Studio. Skills can trigger Studio flows that start with a “Start manually” or “On a schedule” step. A skill can also be dropped into a flow as an “Ask Gemini with skills” step, so an automated email inherits a communication skill rather than a one-off prompt. In the other direction, a skill can extract details from an open email or document and pass them into a manual-start flow — a bridge from an in-app instruction to a runbook, without leaving Gmail to hunt for a chatbot.

Google’s examples are everyday, which is the point of a native interface. A marketing or sales team can keep approved brand voice across Gmail drafts, Docs proposals, and Slides decks. A finance or legal specialist can standardize contract or invoice reviews against guidelines stored in Drive. A project lead can generate a recurring status update from meeting notes, Chat threads, and Drive files without retyping formatting instructions each week.

Sharing is copy-based, not live-linked. A skill shared from Workspace Studio gives the recipient an independent copy; later edits by the author do not overwrite copies already saved. If the skill points at Drive files, those files must be shared separately. Anyone with the share link can see the creator’s email address. Only turned-on skills can be shared. The design is closer to “here is a document you can adapt” than to a centrally enforced corporate standard, and administrators will need to decide whether that is a feature or a governance gap.

Gems give way, but not in one jump

Skills are also the succession plan for Gems in Workspace — with a split that is easy to miss.

Google had already announced that new “Ask a Gem” steps can no longer be added to flows in Workspace Studio. The November 10 post points users to Help Center guidance on replacing Gems with skills in those flows. That is the Workspace Studio half of the transition: the automation builder is moving onto skills.

The Gemini app half is separate, and Google is explicit about the wall between the two. Skills created in Workspace do not sync automatically with skills or Gems in the Gemini app. An upcoming automated migration of remaining Gems in 2027, the company says, will convert Gems to draft skills exclusively in the Gemini app — not in Workspace Studio or Workspace apps. If someone wants the same skill in both places, they will have to copy the instructions themselves.

The chat-box customization layer is not being deleted overnight, and it is not being quietly merged into Workspace. It is being forked. Google is willing to live with two libraries, at least through 2027, rather than force a single skill store across the Gemini app and the productivity suite.

What sources actually say about SKILL.md

Because both of those environments, in Google’s words, “use the open SKILL.md format,” it is tempting to treat that file as a universal standard and stop there. The sources do not quite say that, and it is worth being exact.

What Google says, on the November 10 Workspace Updates post, is three things. First, you can edit Markdown-native (SKILL.md) files directly in Google Workspace Studio. Second, both the Gemini app and Workspace use “the open SKILL.md format.” Third, that shared format is why you can manually copy instructions between the two. Google does not, in that post, define the format further, name a standards body, or claim that every vendor’s “skills” feature reads the same file.

A public specification does exist, independently of Google’s blog post, at agentskills.io. It describes a skill as a directory that must contain a SKILL.md file, with YAML frontmatter that requires a name and a description, followed by Markdown instructions, and with optional scripts/, references/, and assets/ directories. It also describes progressive disclosure: name and description loaded first, the full instruction body loaded when the skill is activated, and extra files loaded only as needed. That is the format the specification itself defines. This article does not treat “open” as a synonym for “implemented everywhere,” and it does not assume that Google’s in-app editor exposes every field or folder in that spec.

Portability, then, is documented as a file convention you can copy by hand, not as a live, cross-product marketplace. A customization that can be opened as Markdown is still a different object from a Gem locked inside one vendor’s chat UI. Whether teams will in fact move skills between Workspace, the Gemini app, and other runtimes is an adoption question, not a claim the current pages can settle.

This is also where the agent-harness layer stays in its lane. A harness is the runtime that loops models, tools, and sessions. A skill is a portable instruction pack that a runtime, or an in-app assistant, can load. Google is making that pack a first-class object in Gmail and Docs. That is complementary to orchestration, not a substitute for it.

What SpaceXSI’s Grok Skills actually offers

SpaceXSI already ships a skills feature for Grok, and it is worth describing it as the company’s own page does, no more and no less.

On Skills in web, iOS, and Android, SpaceXSI presents Skills as “persistent expertise for Grok.” The promise is specific: teach Grok something once and it remembers across every conversation, so preferences, formatting rules, or workflow steps do not have to be repeated. Every Grok account, the page says, ships with built-in skills and no setup required; a user can override a built-in skill with their own version, which always takes priority. Built-in capabilities highlighted there include generating production-ready Word documents, PowerPoint decks, Excel files, and PDFs, and uploading existing files for Grok to restructure or expand.

Custom skills are created in any of three ways the page names: describe a skill in conversation, upload a file, or write one from scratch. A user who has just perfected a format or workflow can also ask Grok to save what it learned as a skill. The launch surface is also named precisely: “Skills are live now on Grok 4.3 across grok.com, iOS, and Android.”

That is the documented product. The official page does not mention SKILL.md, does not describe a Workspace-style in-app @ mention, and does not list desktop, API, or other platforms beyond grok.com, iOS, and Android. Persistence is described as memory across every conversation, not as a file that syncs into Gmail or Docs. This article therefore does not claim that Grok Skills uses the open SKILL.md format, or that a skill built in Grok can be pasted into Workspace Studio, unless and until SpaceXSI says so.

The comparison that is fair is about surfaces. Grok’s skills live in Grok’s own chat clients, with a strong emphasis on producing office documents from that chat. Google’s skills are being installed into the productivity apps themselves, so the customization arrives in the Gmail draft or the Slides deck rather than in a side conversation that has to be copied over. Both are answers to the same user complaint — stop making me re-explain the house style — applied at different layers of the interface.

An interface shift, not a new work OS

It would be easy to inflate this rollout into a claim that Gemini has become the operating system for knowledge work. That is a different article, and we already wrote it. In From Chatbots to Work OS, the question was whether the assistant becomes the place work starts and ends, with connectors and write-capable tools behind it. In The Agent Harness Era, the question was which runtime coordinates models, tools, and sessions. Skills in Workspace sit below both of those claims. They are a customization interface: a way to pin expertise onto the surfaces employees already have open, and to invoke that expertise with an @, an automatic match, or a Studio trigger.

That narrower reading is also more useful. A native skill in Gmail will succeed or fail on ordinary interface tests. Does automatic matching pick the right skill, or a wrong one that rewrites a tone-sensitive email? Do two @ mentions compose cleanly? Can a team treat a shared skill as a template without believing it is a centrally updated standard? Will people maintain two copies — one in the Gemini app, one in Workspace — until the 2027 Gem migration, and what happens when those copies drift? And when will Sheets and Meet, and languages other than English, get the same controls?

The chat box taught users to customize an assistant by talking to it in its own room. Google is now putting that customization on the work surface, in a file format it calls open, with an invocation model that looks less like a conversation starter and more like a mention, a rule, or a scheduled flow. SpaceXSI, for its part, has shown that persistent skills can live inside a general assistant and still be good at emitting documents. The next year will show whether those two ideas stay in separate houses, or whether the SKILL.md file Google is willing to copy by hand becomes the handshake between them.

Until then, the honest description is the modest one. Skills have left the chat box — in Workspace, at least — and become part of the app interface. That is already a platform change.

Related reading

References / Sources

  1. Google Workspace Updates – Scale customized AI across your team with skills in Google Workspace and Workspace Studio (Nov 10, 2026): https://workspaceupdates.googleblog.com/2026/11/skills-in-workspace.html
  2. Google Workspace Studio Help – Learn how skills work in Google Workspace & Google Workspace Studio: https://support.google.com/workspace-studio/answer/17307546
  3. Agent Skills – Specification: https://agentskills.io/specification
  4. SpaceXSI – Skills in web, iOS, and Android: https://x.ai/news/grok-skills
  5. AndBeyond.Tech – From Chatbots to Work OS (internal): /chatbots-to-work-os/
  6. AndBeyond.Tech – The Agent Harness Era (internal): /agent-harness-era/
Share this article

Images: original illustrations generated for AndBeyond.Tech; no logos or real people depicted.