Google LLC’s Cloud division today introduced Gemini agent, a unified artificial intelligence assistant that can act autonomously, generate code and complete work across any device, including web, mobile and desktop.
The new agentic AI experience can be accessed anywhere and through any channel, including the command line, Google Workspace, Microsoft 365 or Slack. It even works within third-party apps and requires no dedicated interface. It is designed to be “omnipresent” and work wherever the user is.
Google Cloud Chief Executive Thomas Kurian said Gemini can answer questions, handle knowledge work, create images and media or take ideas and turn them into finished products.
“You give it objectives, not instructions. You delegate an outcome and come back to finished work,” he explained. “For an agent to do that, it has to be connected to your personal workflows, your systems of record, and your enterprise controls.”
The new agentic experience also follows a common trend in the AI industry: persistent execution. That means it maintains a single set of memories, context and personalization across all communication channels with all its sub-agents. It acts as a teammate that remembers everything the user tells it and can spin out smaller coworker agents with their own dedicated identities and expertise, with their own @agents.company.com emails and their own persistent storage. They also have access only to the context the user or team members provide.
Google Cloud added that the Gemini agent provides flexibility in model choice. Each job runs on a model fitting the task at hand. Simple tasks might run on Gemini Flash, a small, quick model for simple day-to-day tasks or a flagship frontier model like Argon for long-horizon work. Anthropic PBC Claude models are also available today, with other leading private and open models coming later.
Under the hood: skills and tools
Kurian leaned into Google’s vision of providing a full understanding of business context, everything from pricing to product portfolios to departmental norms. Gemini observes and contains the breadth of institutional knowledge that makes a company unique and brings it to the forefront, grounding answers through company data.
Gemini securely connects through collaboration tools such as Confluence, Microsoft Office, Teams, Slack and Workspace. It also works through tools such as Git and Jira, enterprise platforms including Salesforce and ServiceNow, databases such as BigQuery, Databricks, Postgres and Snowflake, and files on desktops.
Users can also build reusable skills that act as instructions, knowledge and workflows that act as prompts that teach agents how to perform specific, multistep tasks. Gemini ships out of the box with a global library of skills, but users can add and publish their own custom skills to a shared company registry, and can build custom skills by asking an agent to take a task and convert it into a reusable template.
The system also learns from every question it is asked and every objective. Session memory, even when it runs for days, talks to people and works with other agents, as it reads documents. It also continues to update its understanding of how users work with it and what it should produce. Google said it spends as much time learning what users are doing as it does using tools.
Google added that it has provided specialized skills built for specific domains, starting with data. For example, the company is extending Gemini with machine learning skills and tools for data scientists and engineers. Describing an outcome in plain language, and Gemini can generate PySpark code, provide a notebook to edit and test it, train a model and troubleshoot issues on its own.
For business users, Gemini can use operational BigQuery reporting skills and the Knowledge Catalog to construct and save queries. This allows the creation of real-time operational reports simply by asking. Once saved, teams can run reports without incurring additional token costs, producing consistent, verified answers every time.
Cost controls at the forefront
Google Cloud fore fronted that although per-token prices dropped significantly since 2024, by about 98%, enterprise AI volume has greatly increased.
Google announced flexible spending options in August and those are coming online today.
Internally, multimodel orchestration designed to pick the right model for the job, or break apart complex jobs across quick models for small tasks and frontier models for difficult parts, reduces costs further. Within this, smart routing automatically triages enterprise workloads so each one runs on the model that delivers maximum performance at the lowest possible cost.
Finally, users can set hard limits on AI spend within the Cloud Billing Console. Gemini will enforce this by monitoring token usage, and if a spend cap is triggered, the agent pauses. The user can resume work in the console with approval, but this requires accepting that it will exceed the set budget cap.
Because tracking is per project, companies can charge AI costs back to specific departments. This lets businesses audit and plan budgets by team and department, enabling more granular cost provisioning.