Google Launches 'Gemini Agent': A Universal AI Coworker Built Into Every App

Google Cloud CEO Thomas Kurian unveiled the Gemini agent at the Gemini at Work 2026 conference — a single, persistent AI agent designed to replace the fragmented AI tools workers juggle today. Rather than a chatbot that forgets context between sessions, the Gemini agent maintains four distinct types of memory (session, semantic, procedural, and episodic), runs continuously in the cloud even after a user closes their laptop, and carries the same context whether accessed on a phone, desktop, command line, or inside Microsoft Teams and Slack.

The scale behind the announcement is striking: nearly 500 Google Cloud customers each processed more than one trillion tokens in the past year, roughly 80% of all Google Cloud customers are now using AI products, and nearly 90% of the Fortune 100 use Gemini Enterprise. That adoption has shifted the conversation from experimentation to mission-critical operations. To match that maturity, the new agent is positioned not as a copilot but as a delegate — one that receives objectives rather than step-by-step instructions and returns finished work.

Architecturally, the Gemini agent introduces several notable features. It supports multi-model orchestration, routing tasks across Google's own Gemini model family, Anthropic's Claude models, and eventually other open and proprietary models — selecting the best fit for cost and quality rather than defaulting to the largest model. Shopify already blends frontier models for millions of merchants this way, and PayPal routes 10 million multi-model requests every week. 'Coworker agents' — persistent sub-agents with their own @agents.company.com email addresses, calendars, and Drive storage — can be created by teams and integrated into workflows like any human colleague.

Inside Google Workspace, the agent goes fully inline: it can schedule meetings by inferring attendees from past chat threads, draft a slide deck from a manager's email request with a single click, or serve as a named team member visible in document version history. Outside Workspace, it connects to Confluence, Jira, Salesforce, ServiceNow, Snowflake, Databricks, GitHub, and any Model Context Protocol server. A company-wide tools and skills registry lets departments publish reusable workflows for others to invoke.

Customer results put concrete numbers on the pitch. Bradesco cut document review time from one hour to five minutes and reduced risk inconsistencies by 60%. Santee Cooper expects 75% faster financial scenario modeling against a $10 billion grid expansion budget. SOMPO built over 10,000 custom agents across 34,000 employees and slashed new model development time from one week to one day. Ulta Beauty tripled digital sales conversion rates using a conversational shopping assistant across 30,000 products. Wesfarmers' Bunnings unit saved staff 500,000 hours of administrative work, while Kmart and Officeworks saw up to 3x higher conversion rates from shopping agents. Orange Spain deployed more than 1,000 custom agents across HR, IT, sales, and customer service teams.