Google Cloud launches the universal Gemini agent: one agent for questions, knowledge work and code, running each job on the model that fits best

What it is. Per the official blog post, Gemini answers questions, handles knowledge work, creates images and media, and writes and runs code, all in a single agent and a single API; the framing is that you give it objectives, not instructions, and delegate an outcome rather than a task. It can work as a personal assistant, or as a team member — a project manager inside a team, for instance, or on behalf of a specific role in an organization, such as an analyst in the finance department. Two of the official examples are concrete: if your manager emails asking for the latest project update as a slide deck, Workspace Intelligence recognizes that as a delegatable task and offers a single-click option to pass it to Gemini; and you can ask Gemini to set up a meeting with the usual team of regional event leads next week without supplying a single name or email address, because Gemini works out who those people are from the membership of your chat space and the thread from your last event, checks their calendars and starts an email thread to coordinate a time, even with external participants.

The six architectural principles Google lists. First, a unified agent: asking questions, working autonomously on objectives you assign, and generating code all happen in one interface, and you can assign work, schedule tasks or have it respond to events. Second, omnipresent access: web, iOS and Android mobile devices, Windows and Mac desktops, and any channel including command line, Google Workspace, Microsoft 365 and Slack; it can also be integrated into third-party applications and surfaces and run as a headless agent that needs no dedicated user interface. Third, persistent execution: because it runs in the cloud it maintains a single set of memories, context and one personalization graph no matter which device or channel you use, so work that takes hours or days keeps running after you close your laptop and is still there when you come back. Fourth, multi-agent orchestration: Gemini can dynamically create a roster of sub-agents — temporary, job-specific agents each with their own identity — to tackle multi-step tasks and coordinate parallel and sequential workflows, and it can also act as a coworker agent, a more team-member-like role with a persistent defined function across days and sessions; coworker agents get dedicated identities including their own @agents.company.com email addresses and their own persistent storage, and only see the context you or your teammates provide. Fifth, deeply contextual: it arrives already knowing your tools, data and work history, learns from every interaction, and teams can create dedicated projects to further tune the context, skills and tools it uses. Sixth is model choice, covered next.

Model choice is the most interesting part of the design. Google's framing: Gemini is the agent, and the model underneath it is a separate choice. It runs each job on the model that fits best — orchestrating across its own Gemini family of models and Claude models from Anthropic today, with other leading private and open models in the future — to deliver optimal quality and lower cost. The stated reason is blunt: the best model for the task is not always the largest one, and matching the model to the work raises accuracy on difficult jobs while lowering cost on simple ones. Google adds a practical point: the leading model changes every few months, so keeping that choice open means your context, skills and data stay put when it does. On cost control Google lists three things: multi-model orchestration to pick the right model for the job, Smart Routing that automatically triages workloads so each runs on the model delivering maximum performance at the lowest possible cost, and real-time project-level spend caps in the Cloud Billing Console. On security and governance it cites identity and policy management, authorization and permission controls, secure sandboxing and network gateways.

Where you can use it. Per the official blog, Gemini works directly inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, carrying the same memory, skills and controls it has everywhere else, and you can mention @Gemini to invoke it in Gmail, Docs, Sheets, Slides and Chat. Device and channel coverage spans web, iOS and Android, Windows and Mac, plus Google Workspace, Microsoft 365, Slack and the command line, and an API lets it be integrated into third-party applications and run as a headless agent. On the customer side, Google lists examples including Every Cure, Harvey, OpenEvidence, Recursion and Replit, plus SMB users of Gemini Enterprise such as Aerotech, EDSpan Global, Pinch Creative, Quadrant Health Group, Salt Productions and Strategem, saying their usage of Google Cloud AI tools grew more than fivefold year over year. A separate SMB-focused post is more practical about integrations: Gemini connects to systems such as Asana, Box, Clay, Docusign, Dropbox, GitHub, LegalZoom, Notion, Salesforce, Shopify, Slack and Wix, and uses context from those tools, your business data and your work history to get more helpful the more you use it.

Three things worth noticing. First, availability is not wide open. The official post says Gemini is in early access with SMB customers today and will be available to all customers soon, while third-party coverage describes it as being in private preview and rolling out first to Workspace customers on select Business and Enterprise plans — a difference in framing whose practical answer depends on your seat type. Second, this is not the same story as Google narrowing free Gemini to a single model on October 9: that one is a consumer plan change, this one is an enterprise agent product, and reading them together leads to the wrong conclusion. Third, universal agent is a launch positioning rather than a finished verdict. Google also put three commitments in the post — taking the friction out with a future-proof platform that runs anywhere and supports your model choice; eliminating complexity by building and integrating the whole stack, silicon through security; and keeping what you build yours, with no need to move people and data and with proprietary inputs and outputs remaining entirely your own. Those three matter more than the feature list, because they decide migration cost and data ownership.