Actualités

Google Cloud unveils persistent Gemini Agents for long-running tasks, and they get their own Gmail, Calendar, and Drive storage

· VentureBeat

Google Cloud has introduced a revolutionary approach to enterprise AI with its new persistent Gemini agents, marking a significant shift from traditional chatbot interactions to autonomous digital workers capable of handling complex, multi-day assignments. These agents represent Google's ambitious attempt to create a unified AI platform that can research information, generate documents, write code, and coordinate other AI systems while maintaining consistent memory and business context across different applications and devices. The standout feature of these Gemini agents is their ability to function as genuine digital coworkers within existing organizational structures. When configured for team use, these agents receive their own corporate identities complete with Gmail addresses, Calendar access, Drive storage, and directory presence. This allows employees to delegate work to AI agents much like they would assign tasks to human colleagues, including mentioning them in Google Chat or requesting document modifications. Unlike previous AI assistants that required users to manage separate tools for different functions, Google's new system operates through a single interface that can work across Google Workspace, Microsoft 365, Slack, desktop applications, mobile devices, and command-line environments. The architecture supports multiple AI models beyond Google's own Gemini, including Anthropic's Claude, with plans to incorporate additional proprietary and open-source models. The timing of this announcement reflects the intensifying competition in enterprise AI. Microsoft recently updated its Copilot platform with persistent automation capabilities, OpenAI launched always-on agents, and Anthropic introduced native Claude integration with Google Workspace applications. This competitive landscape demonstrates the industry's collective push toward AI systems that function as continuous digital workers rather than reactive chatbots. Google's agents utilize a sophisticated memory system encompassing session memory for ongoing tasks, semantic memory for accumulated knowledge, procedural memory for learned workflows, and episodic memory covering previous work history. This comprehensive approach enables agents to develop institutional knowledge about company processes and employees, reducing the need for repeated explanations. Security and access controls remain paramount in the enterprise environment. Google implements cryptographically attested identities for each agent, operates under existing organizational security policies, and restricts agent access to explicitly shared files through Workspace permissions. All agent activities are logged for auditing purposes, and the system includes safeguards to prevent unauthorized transmission of sensitive information to external AI models. The new Gemini agents are currently available in private preview to select customers, with broader availability promised soon. They will be integrated directly into Gemini Enterprise at no additional cost for existing Google Workspace business and enterprise subscribers. This strategic positioning could significantly impact the enterprise AI market by making advanced agent capabilities accessible to organizations already using Google's productivity suite, potentially accelerating adoption and setting new standards for workplace AI integration.

Google Cloud has launched a groundbreaking enterprise AI initiative with persistent Gemini agents that fundamentally reimagine how artificial intelligence integrates into workplace operations. These advanced agents transcend traditional chatbot limitations by functioning as autonomous digital workers capable of managing complex, extended assignments while maintaining consistent identity and context across multiple platforms and timeframes.

The announcement, made at Google's Gemini at Work 2026 event, positions these agents as comprehensive workplace solutions rather than specialized tools. Unlike previous AI assistants that required users to coordinate multiple separate applications, Google's unified approach allows employees to delegate entire objectives to a single agent that determines execution strategies, accesses necessary enterprise systems, and delivers completed work.

A particularly innovative aspect involves configuring agents as persistent digital coworkers with full organizational integration. These AI entities receive complete corporate identities including dedicated Gmail addresses, Calendar access, Drive storage allocation, and company directory listings. This infrastructure enables natural workplace interactions where employees can assign tasks, mention agents in group communications, and collaborate as they would with human colleagues.

The technical architecture demonstrates remarkable flexibility in both deployment and model selection. While Google provides its own Gemini models as the foundation, the system already supports Anthropic's Claude models with plans to incorporate additional proprietary and open-source AI systems. This model-agnostic approach could prove strategically important as organizations seek to avoid vendor lock-in while leveraging best-in-class capabilities for specific use cases.

Google's timing reflects the rapidly evolving enterprise AI landscape. Microsoft recently enhanced its Copilot platform with persistent automation features, OpenAI introduced always-on agent capabilities, and Anthropic launched native integration with Google Workspace applications. This competitive intensity underscores the industry's recognition that the future of workplace AI lies in continuous, autonomous operation rather than reactive assistance.

The memory architecture represents a sophisticated approach to institutional knowledge management. By implementing session memory for active tasks, semantic memory for accumulated information, procedural memory for learned workflows, and episodic memory for historical context, these agents can develop deep understanding of organizational processes and relationships. This comprehensive memory system reduces training overhead and enables more contextually appropriate responses over time.

Security considerations receive substantial attention given the sensitive nature of enterprise data. Google implements cryptographically attested identities for each agent while maintaining compatibility with existing organizational security frameworks. Access controls limit agents to explicitly shared files through standard Workspace permissions, and comprehensive logging ensures full audit trails for all agent activities. Additional safeguards prevent unauthorized transmission of sensitive information to external AI models, addressing a critical concern for organizations handling confidential data.

The integration strategy emphasizes seamless adoption by working within familiar environments including Google Workspace, Microsoft 365, Slack, desktop applications, mobile devices, and command-line interfaces. This approach eliminates the need for employees to learn new interfaces or modify existing workflows, potentially accelerating organizational acceptance and reducing implementation friction.

Google's decision to include these capabilities at no additional cost for existing Gemini Enterprise customers could significantly impact market dynamics. By making advanced agent functionality accessible through current Google Workspace subscriptions, the company positions itself competitively against standalone AI platforms while potentially expanding its enterprise footprint.

The broader implications extend beyond individual productivity gains to fundamental changes in how organizations structure work and allocate resources. As AI agents become capable of handling multi-day projects, coordinating with other systems, and maintaining institutional memory, traditional boundaries between human and artificial intelligence contributions may continue to blur, requiring new approaches to management, accountability, and organizational design.