Google’s Gemini Agent Wants to Be Your New Coworker

Your daily brief on leadership and business

Google Cloud has announced the Gemini agent, a single universal assistant for work that answers questions, does knowledge work, creates images and media, and writes and runs code, all from one prompt box. Google Cloud CEO Thomas Kurian summed up the idea in his keynote: work now starts in the prompt window. Instead of step-by-step instructions, users hand Gemini an objective, and the agent plans the job, picks the tools and returns finished output inside the apps people already use.

A colleague with its own inbox

The most eye-catching feature is the “coworker agent”. A user describes a role, and Gemini creates an agent with its own Workspace account, email, calendar and Drive. It acts under its own identity, sees only what is shared with it and shows up under its own name in version history, so at least nobody can blame it for edits it did not make. For multi-step jobs, Gemini can spin up temporary sub-agents that work in parallel or in sequence for hours or even days, and long-running work keeps going after the user closes the laptop.

Inside Google Workspace the agent works in Gmail, Docs, Sheets, Slides, Chat and Calendar. Google’s example: it can schedule a meeting with a regular team without being told names or emails, by finding attendees from chat spaces and past events and checking calendars. A feature called Workspace Intelligence spots delegatable requests, such as a manager asking for a slide deck, and offers a one-click handoff. Gemini also connects to tools such as Slack, Jira, Confluence, Salesforce and Microsoft Office, and it can route jobs between Google’s Gemini models and Anthropic’s Claude models.

Guardrails and a spending brake

Google frames governance around four questions: who the agent is, what it may do, what it did and what it must never touch. Every agent gets an attested identity with least-privilege permissions, every action is logged to the agent rather than a person, and agents run in a sandbox behind a gateway that enforces company policy. For finance teams there are real-time spend caps per project: when a cap is hit, the agent simply pauses until someone resumes it with one click. Kurian noted that per-token prices have fallen 98% since 2024, while enterprise AI volume has grown fast.

Google also listed customer results, including Commerzbank cutting document checks from 20 hours to one and Snap reducing diagnostic troubleshooting from 30 minutes to 30 seconds. For leaders, the message is clear: the productivity pitch has moved from “AI that helps you write” to “AI that takes the task off your plate”, and the new management question is who reviews the work of the colleague that never sleeps.