At its Gemini at Work event on October 8, 2026, Google Cloud unveiled the Gemini agent: one universal work agent that answers questions, performs knowledge work, generates media, and writes and runs code from a single prompt box. The real story is the architecture — an agent that creates synthetic coworkers with their own identities, and that routes tasks to Anthropic’s Claude as well as Google’s own models.
Thomas Kurian, Google Cloud’s CEO, led the launch in a keynote-adapted post: work now starts in the prompt window.
- One interface, one API: chat, autonomous execution and coding in a single agent.
- Sub-agents on demand: temporary, identity-bearing helpers that split long jobs into parallel steps.
- Multi-model orchestration: tasks route to Gemini or Claude, with Smart Routing and real-time spend caps.
One agent, every device, one memory
Architecturally, Gemini is unified: chatting, autonomous work and coding share one interface and one API. Customers assign outcomes instead of step-by-step instructions, and the agent keeps running in the cloud after a laptop closes. It is reachable from the web, iOS, Android, Windows, macOS, the command line, Workspace, Microsoft 365 and Slack, carrying one shared memory across every surface.
“You give it objectives, not instructions. You delegate an outcome and come back to finished work.” — Thomas Kurian, Google Cloud
Connectors reach Confluence, Teams, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Snowflake and any MCP server. Kurian describes four memory types — session, semantic, procedural and episodic — so the agent retains the current task, what it has read, how a job is done and everything it did before.
Coworkers with inboxes and disposable sub-agents
Gemini does not work alone. It creates temporary sub-agents, each with its own identity, to split long jobs into parallel or sequential steps, and it can act as a coworker agent — a stable teammate whose defined role persists across days and sessions. Coworkers receive dedicated identities, including addresses in the form @agents.company.com, plus their own storage and, inside Workspace, an account, calendar, Drive space and directory presence. Their edits appear under the agent’s name in version history.
Google stresses that a coworker sees only what a team shares with it, and that every action is logged under the agent’s identity rather than a person’s. That is a design promise, not a verified guarantee: no external audits or error data accompanied the launch. The leap from Gemini Spark, Google’s personal agent, is one of scale — Spark served an individual, this targets the org chart.
Why Google routes its own agent to Claude
Google separates the agent from the model: Gemini orchestrates, while the model underneath is chosen per task. Today that means the Gemini family plus Anthropic’s Claude, with more models promised later. The logic is economic — the best model for a job is not always the biggest, so matching them lifts accuracy on hard tasks and trims cost on simple ones. Smart Routing and real-time spend caps per project add a hard stop: hit the ceiling and the agent halts until a human resumes it.
Volume does the rest of the talking. Kurian says nearly 500 Google Cloud customers each processed more than one trillion tokens in the past year, and close to 90% of the Fortune 100 uses Gemini Enterprise. Almost 80% of all Google Cloud customers now run the company’s AI products, while the price per token has fallen 98% since 2024.
Customers, caveats and the turf war ahead
The customer roster includes a Spanish deployment: Orange Spain has rolled out more than 1,000 custom agents, built by HR, IT, sales and customer-service staff to triage incidents and analyse network problems. BNP Paribas embedded the agent in an internal assistant for over 65,000 employees, and PayPal routes 10 million multi-model requests weekly. All figures are Google’s or its customers’, so they measure ambition rather than audited results.
What the announcement omits matters too: no price, no general availability date. Sector-specific versions arrive in preview for financial services and legal, with government, healthcare and commerce promised soon. The timing is pointed — the same day, Satya Nadella published an essay casting Copilot as a new operating system for work. Google and Microsoft now say the same thing in different words: the agent is the front door, and whoever owns it owns the software behind it.
The most consequential part of Gemini at Work 2026 is not the universal agent but the coworker: a synthetic employee with an account, a calendar and a signature in the version history. That is an organisational shift, not a product upgrade, and few enterprises have policy for it yet. Until Google publishes pricing and a rollout calendar, the announcement remains a well-argued statement of intent.