Can AI agents automate Google Docs?

Google Docs now has first-party agent infrastructure as well as a mature document API. The REST API creates documents, returns structured text and formatting, and applies ordered atomic batch updates for text, styles, lists, tables, images, headers, footers, tabs, and other supported document elements. Revision-aware write controls let clients reject or merge around concurrent collaborator edits, but index-based mutations still require a fresh read and careful reverse-order editing. Some editor features and the new comments and suggestions API remain outside stable full coverage, while import and format export use the Drive API. OAuth user consent is the normal access path. Service accounts can work with directly shared documents or, under super-admin-approved domain-wide delegation, act for Workspace users. Google publishes client libraries in six common languages. Its new `gws` command-line client dynamically exposes the Docs discovery surface with structured JSON, schema introspection, dry-run support, and an append-text helper. Google also operates a remote Docs MCP server with `read_doc` and `update_doc`; it is official and permission-aware, but still a Developer Preview with only two tools, so it is not yet a production-complete MCP path. Google explicitly warns that document content can cause indirect prompt injection and says users should review agent actions. Native Docs webhooks, current third-party workflow-connector quality, and repeatable computer-use viability were not established.

Best path today: api · Overall automatability: 8/10