A Universal Agent, Not Another Chatbot
On October 8, 2026, at its Gemini at Work 2026 event, Google Cloud unveiled a new universal AI agent designed to take on entire assignments rather than respond to individual prompts. The agent can research information, create documents, write code, coordinate other AI agents, and complete tasks lasting hours or days, all while maintaining the same memory, business context, and identity across different applications and devices.
The agent is available starting today in private preview to a select group of customers, with wider availability promised soon but no specific date given. Once broadly released, it will be built directly into Gemini Enterprise at no additional charge for customers where Gemini Enterprise is available, including through existing Google Workspace business and enterprise subscriptions.
The agent works across Google Workspace, Microsoft 365, Slack, desktop and mobile applications, and the command line. Google likened its deployment to a software version update.
AI Coworkers Get Their Own Email, Calendar, and Directory Entry
The coworker configuration is the most notable feature. Google intends to give these agents their own identities within an organization's existing productivity infrastructure.
A manager could create a Gemini Event Planner Agent that receives its own Workspace account, email address, calendar, Drive storage, and company-directory presence. Employees could then assign work to the agent much as they would to a colleague, including by mentioning it in Google Chat or requesting document edits.
Because execution takes place in the cloud, employees can start an assignment on one device, leave it running, and resume from another without reconstructing the context. The agent can manage workflows lasting multiple days, coordinate parallel assignments, and respond to scheduled tasks or events. It can also assemble temporary groups of specialized sub-agents for complex work, distributing different responsibilities among them.
Multi-Model Routing: Gemini Plus Claude
The architecture is notably flexible when it comes to AI models. While Google supplies its own Gemini models, the agent can already use Anthropic's Claude models, with support for additional proprietary and open-weight AI models planned.
This multi-model approach means the agent is not locked into a single AI provider. Smart Routing determines which model handles each task, and Google has introduced real-time spend caps that pause agents when budget limits are reached. According to PPC Land reporting on October 8, 2026, per-token prices have fallen 98% since 2024, yet enterprise AI volume keeps straining budgets, making spend controls a practical necessity.
The agent also supports a headless mode for API-only operations, with no chat interface. This allows it to run background workflows, process data pipelines, and handle automated tasks without human interaction.
Memory Architecture: Four Types of Context
The agent's memory spans four categories:
| Memory Type | What It Stores | Business Use |
|---|---|---|
| Session memory | Ongoing task context | Resuming work across devices |
| Semantic memory | Accumulated knowledge | Company processes, institutional info |
| Procedural memory | Learned workflows | Repeating tasks without retraining |
| Episodic memory | Previous work history | Learning from past assignments |
This architecture is designed to help the agent develop an understanding of a company's processes, employees, and institutional knowledge instead of requiring users to explain the same information repeatedly. To connect with business operations, the agent supports Salesforce, ServiceNow, Jira, Git, BigQuery, Snowflake, Databricks, and Microsoft Teams. Enterprises can also connect Model Context Protocol (MCP) servers and publish reusable instructions to shared internal registries.
Security: Cryptographic Identity and Isolated Execution
Google says each agent receives its own cryptographically attested identity and operates under the organization's existing security policies and permissions. Complex tasks execute inside isolated containers with restricted outbound communications. An Agent Gateway applies company security rules to traffic between agents and external systems.
Agents can access only files explicitly shared with them through Workspace permissions. Google's security controls proactively stop workflows that attempt to send sensitive information to unauthorized AI models, including third-party offerings. Every agent action is logged for auditing, with activity associated with the agent's own identity while retaining accountability through enterprise administration controls.
However, questions remain. Google has not specified whether information is transmitted to external model providers for processing when using third-party models like Claude, or detailed the data retention and residency arrangements applicable to those models. This distinction matters for organizations deploying agents across sensitive financial records, customer databases, and legal documents.
How It Compares With Competitors
Google's announcement arrives amid intensifying competition to provide the central AI interface for enterprise work:
| Company | Product | Key Differentiator | Date |
|---|---|---|---|
| Gemini agent | Own email, multi-model, spend caps | Oct 8, 2026 | |
| Microsoft | Copilot (Home, Code, Autopilot) | Autopilot background agent, Frontier program | Sep 25, 2026 |
| OpenAI | Dots agents | Always-on, cross-app, Spaces | Sep 29, 2026 |
| Anthropic | Claude for Google Workspace | Edits Docs, Sheets, Slides directly | Oct 6, 2026 |
| xAI | Grok Bot / Team Bots | Shared team agents | Aug-Sep 2026 |
Each approach reflects a different philosophy. Google bets on persistent identity and deep Workspace integration. Microsoft unifies productivity, development, and automation. OpenAI focuses on continuous autonomous work. Anthropic embeds into existing tools. The common thread: AI companies increasingly want their assistants to become long-running digital workers, not chatbots that respond to prompts.
What This Means for Businesses
For companies evaluating AI agents, several practical considerations emerge:
- Agent identity is a new attack surface. An AI coworker with its own email and calendar can be phished, spoofed, or socially engineered. Security teams need to treat agent identities with the same rigor as human ones.
- Spend management is now a feature, not an afterthought. Google's real-time spend caps address a real pain point. Without them, autonomous agents running for days could generate significant costs.
- Multi-model routing changes vendor strategy. The ability to use Claude inside a Google product means companies are no longer locked into a single AI provider. Procurement and governance need to account for multi-model environments.
- Organizational design shifts. When an AI agent has a directory entry and a calendar, it becomes part of the org chart. Companies need policies for agent provisioning, deprovisioning, and scope of authority.
- Private preview means preparation time. Most organizations cannot deploy this yet. Use the lead time to define use cases, establish governance, and identify which workflows benefit most from persistent agent identity.
Read Next
- The 2026 AI Trust Report: Which Brands AI Actually Trusts — 3.9 million citations reveal AI trust patterns
- How QuickBooks Earned a 223% Citation Lift in AI Search — A concrete GEO playbook
- Agentic Payments Fraud: Blind Spots in Fraud Controls — When AI agents make payments, fraud detection needs to evolve
- Backbase Conversational Banking: Agentic AI Goes Live at 10+ Banks — Agentic AI in production with governance
