Build2 publishersIndependently confirmed2 min readPublished
Claude's add-on for Docs, Sheets and Slides lets each user turn off edit approval
Anthropic's Claude beta for Google Docs, Sheets and Slides, announced October 6, lets each user swap "Ask before edits" for "Accept all edits". That puts the review step on shared company files in the hands of whoever opens the sidebar.
The Engineer · Build desk
Drafted by a language model from the sources cited here and checked against its claim ledger before publication. How we use AISend a correction
What happened
- The add-on is in beta on every paid Claude plan, Pro, Max, Team and Enterprise, and installing it requires permission to add Google Workspace Marketplace apps.
- Workspace admins can deploy it across an organisation, limit it to selected groups or block Marketplace apps, and a user without that allowance cannot install it alone.
- According to Anthropic's support documentation, the add-on reads only the open file, and other Drive files stay out of reach unless the user enables a connector.
- Separate Docs, Sheets and Slides connectors let users start in Claude with a pasted link or a new file, with access Anthropic says follows the user's Google sharing permissions.
- Anthropic says larger rewrites appear as suggestion cards to apply or dismiss, while small changes such as fixing a sentence or a heading can be made directly in the file.
Compiled by The EngineerSomething wrong?How this is made
Why it matters
- decision Approving the add-on for a group also means accepting each member's choice between "Ask before edits" and "Accept all edits" on shared files, since the report places that switch with the user.
- exposure Reviewing an AI edit in a shared file comes down to the diff: collaborators can find the change in version history, but the prompt behind it stays in one user's browser.
- constraint Pro and Max subscribers on a managed Google account still wait on their Workspace admin, because the Marketplace permission gate applies on every paid plan.
The add-on opens in a sidebar beside the active file. It reads the file, takes the user's selection into account and writes changes in place [15]. By default, that path includes a review step [3]. Under "Ask before edits", Claude presents its proposed changes for approval [2]. I think that is the right default. A user can switch to "Accept all edits", and Claude then applies changes without stopping [2]. The suggestion-card split is a second threshold, and Anthropic sets it [4]. Runtimewire's report does not say where that threshold sits, or whether a Workspace administrator can lock the approval mode for a group.
Scope is also a user decision. According to Anthropic's support documentation, turning on a connector is what takes Claude past the open file [8]. From there, its reach follows the user's own Google sharing permissions, Anthropic says [7]. As the report describes it, the administrator's lever comes earlier, at the Marketplace install permission [6].
The hard limits are good engineering. Anthropic's documentation lists actions Claude cannot take, including changing file ownership or sharing access [10]. A run in "Accept all edits" mode can rewrite what a shared deck says. It cannot change who owns that deck or who can open it [17]. Runtimewire notes that in shared company documents, an AI edit has consequences beyond the person who prompted it [14].
Google already ships Gemini in all three apps, with a side panel for help with files [11]. Runtimewire argues the integration matters more as a distribution move than as a new document editor [12]. Its case is that some users want Claude's models and their existing Claude connectors while keeping their work in Google's file formats [13]. The Sheets capabilities, including formulas, pivot tables and Python for some data-cleaning and joining tasks, are Anthropic's own product claims [18].
What to watch
- An admin setting that forces "Ask before edits" for a domain or group would move the edit policy from each user's sidebar to the Workspace console.
- Documentation of where Anthropic draws the line between a suggestion-card rewrite and a direct small edit, and how each approval mode handles the two.