Google Cloud introduced the Gemini agent Thursday, an AI system that plans and carries out multistep work across corporate software and can create sub-agents with their own email addresses and storage, SiliconANGLE reported. Google unveiled it at its Gemini at Work 2026 event and is offering it to businesses before consumers.

What Happened

Google describes the product as a universal agent for work. Users start a job from a single prompt, and the agent breaks it into steps, uses tools and connected business systems, and returns finished output into documents, inboxes or developer environments. Google Cloud CEO Thomas Kurian summed up the pitch as giving the agent "objectives, not instructions," according to SiliconANGLE.

The agent runs on the web, mobile apps, desktop and the command line, and inside Slack and Microsoft 365. SiliconANGLE listed connectors for Google Workspace, Microsoft Office and Teams, Jira, Confluence, Git, Salesforce, ServiceNow, BigQuery, Databricks, Postgres, Snowflake and local desktop files. VentureBeat reported that it can also connect to Model Context Protocol servers.

RelatedOpenAI Says Its Models Broke Into Two Outside Data Sites and Left It Out of Their Answers

For larger jobs, the agent can spin up sub-agents. SiliconANGLE said these get @agents.company.com email addresses and can see only the context they are given. TechCrunch reported that an agent can have its own Workspace account with email and calendar, can be emailed or added to group chats like a colleague, and leaves audit trails attributed to itself. VentureBeat described four kinds of memory, covering the current session, accumulated knowledge, learned procedures and past work, with tasks that can run for hours or days.

Model selection is automatic by default. SiliconANGLE said simple jobs go to Gemini Flash and complex work to Argon, Google's frontier model, while users can override the choice and pick Anthropic's Claude models. Google said private and open models will be added later.

Why It Matters

The identity design is aimed at a familiar objection from security teams: agents that act with an employee's full permissions make it hard to tell who did what. Giving each agent its own account, limited access and a separate audit log treats it more like a separate worker than a feature of an employee's account. VentureBeat reported that agents receive cryptographically attested identities and permissions.

Cost is the other pressure point. SiliconANGLE said Google is pairing the launch with hard spending caps enforced in the Cloud Billing Console, which pause agents when a limit is hit, and per-project cost tracking for charging departments. Google said token prices have fallen about 98% since 2024 though enterprise AI volumes have climbed sharply over the same period.

Google also has scale to lean on. Alphabet CEO Sundar Pichai said Gemini has more than 1 billion monthly users and that nearly 90% of the Fortune 100 use Gemini Enterprise, TechCrunch reported. On, Shopify and PayPal tested the agent early, and Gemini Enterprise customers cited by Google include BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty and Wesfarmers.

Context

The launch comes as the largest AI developers try to turn chat assistants into software that works for long stretches inside company systems. TechCrunch named Meta's Muse, ChatGPT's Dots and messaging-based agents such as Instinct as competing efforts.

Reports differ on how widely the agent is available. SiliconANGLE described the features as available now, while 9to5Google said the agent is in private preview and coming soon to Workspace customers on select Business and Enterprise plans.

What to Watch

Google has not given a date for a consumer version or detailed pricing beyond its spending controls, and VentureBeat noted that no independent benchmarks yet show how reliably the agent handles long, multi-application tasks. Also pending are the promised support for private and open-weight models and the timing of general availability for Workspace customers.