Google’s Gemini Agent Gives Marketing Work a New Identity

Google’s new Gemini agent can have its own email address, storage, and persistent work identity. For marketing teams, that turns AI adoption into an access-control decision: what should an agent be allowed to see, change, and send under its own name?

Google announced the Gemini agent at its October 8 Gemini at Work event. The company describes it as a general work agent that can plan tasks, use tools, connect to business systems, and deliver work inside products such as Gmail, Drive, Docs, and Chat. It can also route some jobs to Anthropic’s Claude models. Google says business access begins with an early-access program, so this is a product direction and test, not a broadly available capability today. TechCrunch’s launch coverage confirms Google is starting with business customers.

The most consequential product detail is identity. Google says “coworker agents” can have their own company email, persistent storage, and a defined role across sessions. A team can share specific context with the agent. TechCrunch’s report adds that users can tag the agent, email it, share files with it, or add it to group conversations, with its actions recorded under the agent’s identity. The model fits a longer shift toward agents that change how marketing work gets organized, while creating new questions for marketing operations teams already automating workflows.

Thomas Kurian, Google Cloud’s CEO, described the shift as giving agents “objectives, not just instructions.” The distinction is meaningful for marketers. An instruction might ask for three subject lines. An objective could ask the agent to prepare a launch-readiness document, gather campaign details from multiple systems, and respond to comments on a shared file. Each extra step requires more authority than a blank-page writing assistant.

Google’s example of marketing work involves an event coordinator agent that prepares launch materials and handles document comments. That use case hints at a new job design: marketers may delegate a bounded workflow to an agent, review its work, and be accountable for actions that happen while they are not watching each intermediate step.

There is a useful real-world counterpoint to the product framing. In a public LinkedIn post shared by Retailers Magazine, Santos Daniel Arce Saldívar, executive director of customer experience and innovation at Mexican pharmacy chain Farmacias del Ahorro, described the company’s AI program as serving both customer experience and internal product-catalog work. He said, translated from Spanish, “internally, we focus on creating agents designed to enrich the product catalog within our digital ecosystem.” The same post says the retailer estimates 140,000 hours in annual operating time savings and reports 90% active employee use. Those figures come from the retailer’s account in a LinkedIn post referencing Google Cloud’s case study, not an independent audit.

That example makes the managerial questions tangible. If an agent updates a product catalog, its scope may affect names, descriptions, search terms, or claims visible to customers. If it drafts launch documents, it may need access to product, legal, and campaign files. If it replies to collaborators, it can create commitments or imply approval. Identity makes the audit trail clearer, but it does not itself make permissions appropriate.

The tool footprint is broad. Google says Gemini can connect with Workspace, Microsoft 365, Slack, Salesforce, Jira, databases, and MCP servers. That makes governance more than an IT checklist. Marketing operations needs to define which systems are read-only, which actions require a person, what content an agent can publish or send, and how access is removed when a project ends.

The company also highlights model choice, cost caps, policies, and administrative controls. Those controls matter, but the quality of a workflow still depends on its assignment. Teams should begin with a task map: data needed, decisions the agent may make, external actions it may take, and the human who signs off. Measure accuracy and completion time alongside correction burden and escalations. A faster draft that leaves an employee to inspect every hidden action may not be faster work.

Google’s launch is also a platform strategy. A persistent agent that retains context and works across a company’s systems can become part of the operating layer rather than another app tab. That could reduce tool switching, while making it harder for a company to move its workflows and memory elsewhere. Google says model choice will remain flexible, but customers will need to test what context and execution history travel with them.

The verdict: Treat an agent’s identity like a new service account with a job description. Before assigning marketing work, decide what it can access, which actions it can take, and where a human must approve. The inbox is a feature. The permission boundary is the product decision.