Gemini Agent for Work: What Google's Universal Agent Does

Google has introduced the Gemini agent, a single AI agent that handles research, writing, media creation and coding from one prompt box, and it marks a real shift in how the company wants businesses to use AI. Announced at Google Cloud's Gemini at Work event on October 8, 2026, the agent is designed to take objectives rather than step-by-step instructions, keep working in the cloud for hours or days, and pick the most cost-effective model for each part of a job, including Anthropic's Claude models alongside Gemini.

• One agent, one API: Google describes the Gemini agent as a universal agent for work that answers questions, does knowledge work, creates images and media, and writes and runs code.
• Model choice: it routes jobs across the Gemini family and Anthropic's Claude models today, with other private and open models planned, using a Smart Routing tool to balance quality and cost.
• Long-running work: jobs run in the cloud and continue after you close your laptop, with temporary sub-agents handling parallel steps.
• Rollout: businesses first, consumers later, according to TechCrunch; Google has not announced pricing or a general availability date.

What Is the Gemini Agent?

Most AI assistants still work turn by turn: you ask, it answers, you ask again. The Gemini agent is built around delegation instead. Google Cloud CEO Thomas Kurian framed it as giving the system "objectives, not just instructions," and the agent plans the work, calls tools and skills, and reports progress in a tasks inbox that shows its reasoning and any sub-agents it has spun up, according to TechCrunch.

It is available across the web, iOS, Android, Windows and Mac desktops, the command line, Google Workspace, Microsoft 365 and Slack, and it can be embedded in other apps or run headless.

How Does the Gemini Agent Choose Between Gemini and Claude?

The most notable design choice is model flexibility. Google says the agent routes each job across its own models (Argon for frontier reasoning, Flash for speed and volume, Omni for generative media and Gemma for open-weight edge work) and Anthropic's Claude models. Users can also pick a model manually. A Smart Routing tool triages each task to whichever model delivers the needed quality at the lowest cost.

For enterprises, this is a meaningful step toward multi-model AI as a default rather than a custom integration project.

Coworker Agents With Their Own Identity

Google is also introducing persistent, role-based "coworker agents." Each one gets its own identity, including an @agents.company.com email address and, in Workspace, its own account with calendar and Drive. These agents act under their own name and only see what has been shared with them, so their work is visible in version history and audit logs rather than hidden behind a human's account.

The agent connects to Confluence, Jira, Git, Salesforce, ServiceNow, BigQuery, Databricks, Postgres, Snowflake and local desktop files, plus any Model Context Protocol (MCP) server. Google lists four kinds of memory: session memory for the current task, semantic memory built from documents, procedural memory for how work gets done, and episodic memory of past actions.

Guardrails: Spend Caps, Sandboxes and Audit Trails

Agentic AI raises fair questions about cost and control, and Google answered both directly:

• Hard spend caps: per-project limits in the Cloud Billing Console pause an agent when the cap is reached, and it resumes with one click.
• Attested identity: every agent has a cryptographically attested identity with least-privilege permissions.
• Isolation: agents run inside an Agent Sandbox, and all traffic passes through Agent Gateway, an AI network firewall that enforces company policy.
• Accountability: every action lands in an audit trail attributed to the agent itself.

Industry Editions and Early Results

Google also previewed specialized editions. The Financial Services edition includes more than 50 foundational skills and data from FactSet, LSEG, S&P Global and SEC filings, with CME Group and Deutsche Bank named as users. A Legal edition works with Harvey and Onit, building on Gemini Enterprise for Legal. Government, Healthcare and Retail editions are listed as coming soon.

Google says Commerzbank cut a manual document review from 20 hours to one, and Sundar Pichai said Gemini now has more than 1 billion monthly active users, according to TechCrunch.

Why the Gemini Agent Matters

The industry has spent two years building AI agents one workflow at a time. Google is betting that the next phase looks like a single, governed agent that can use many models, many tools and many surfaces, with cost and security controls built in from day one. If that holds up in real deployments, it could make delegating serious work to AI feel as routine as sharing a document. Follow more developments in our AI news coverage.

Sources: Google Cloud Blog — Welcome to Gemini at Work 2026 — October 8, 2026; TechCrunch — Google brings agentic AI to Gemini, starting with businesses — October 8, 2026; The Register — Google Cloud casts Gemini as your enterprise AI hero — October 8, 2026.