Google's new Gemini agent gets its own Workspace email address and a seat in your inbox

Google began rolling out a single, persistent AI agent inside Gmail, Docs, Sheets and Slides on October 8, and gave it something no Workspace add-on has had before: its own account, its own email address, and a named presence in the tools a marketing team already works in every day.

The move, announced at Google Cloud's Gemini at Work 2026 event and detailed in a post from Google Cloud CEO Thomas Kurian, folds what had been a scatter of separate Gemini surfaces — chat, Gemini Enterprise, Workspace add-ons — into one agent reachable from a single prompt box, delegable by a manager, or added to a document or chat thread as a named coworker. TechCrunch reported that the rollout targets businesses first, with a consumer version to follow, and that every action the agent takes is logged under the agent's own name, not the name of the employee who asked for it.

What actually changed

Three things separate this from last year's Gemini. First, identity: this coworker agent gets its own Workspace account, calendar, Drive folder and directory listing, so it can be @mentioned in a comment, tagged in Chat, or emailed directly, and its edits show up under its own name in document version history. Second, persistence: the agent keeps one memory and one personalization graph across sessions, so work it starts before someone closes a laptop keeps running in the cloud. Third, choice: Google is routing tasks across its own Gemini models and, as of this launch, Anthropic's Claude, with other outside models planned — a tacit admission that locking marketing and ops teams into one model family is no longer how platform vendors compete.

The agent also writes PySpark, builds financial models in Sheets from a plain-language request, and generates on-demand operational reports wired to BigQuery. Saved reports can be re-run later at no added usage charge, Google says. Google is pairing that with real cost controls: a team can set a hard dollar ceiling on what one project's agent work is allowed to spend, and once that ceiling is reached, further tasks pause until someone signs off on more budget.

Who this hits

Any software company running its marketing org on Google Workspace now has a new, non-human teammate showing up in the inbox, not a hypothetical one. That reshapes two groups at once. Marketing operations staff who build the weekly pipeline report or the QBR deck by hand get a tool that can be handed that exact task and will show its reasoning, sources and draft before it finishes, surfaced automatically whenever the system flags a request as something that could be delegated entirely rather than answered in chat. And martech vendors whose product is a reporting layer, a deck generator or a lightweight BI dashboard are now competing with a capability bundled into a productivity suite most of their prospects already pay for.

Bloomberg Media's CTO, William Anderson, is one of the named early users, and said the appeal is less the model and more what it is connected to:

"By grounding our AI in a trusted institutional context, we ensure confidence in the accuracy and quality of every insight generated." — William Anderson, CTO, Bloomberg Media

That framing — an agent is only as useful as the systems it can see — is the part software marketers should sit with. Google lists connectors into Salesforce, ServiceNow, Jira, Confluence, Slack, Snowflake, Databricks and any internal Model Context Protocol server, meaning the agent's reach into a company's actual revenue and campaign data is a procurement and governance decision, not a product feature toggle.

What to do differently

Treat this as an identity and access question before it becomes a tooling question. A coworker agent that can be @mentioned, emailed and granted document permissions is a new account in your identity system, and marketing is usually the department with the loosest list permissions, the widest CRM exports and the most third-party tool connections in the company. Audit what that account could see before anyone asks the agent to pull a campaign report.

• Map what Workspace, CRM and ad-platform data a coworker agent account would inherit by default, and scope it down before granting access.
• Ask whether your own product's reporting or deck-building features are now a commodity Google ships for free, and reposition around the judgment or data you add on top.
• Check vendor contracts for AI-agent seat language now, since Salesforce, HubSpot and Atlassian already price agent access separately from human seats — Workspace is likely next.
• Pilot the cost-cap feature on one project before rolling an agent out company-wide; a runaway task queue is now a budget line, not just a support ticket.