The system can handle complex tasks and utilize Claude models
Google Cloud has announced the launch of «Gemini agent», a universal tool for enterprise customers designed to automate various workflows. The system is capable of utilizing both native Gemini models and third-party solutions, including Claude from Anthropic. This was reported by 9to5Google, according to the portal PromPolitInform.
The new product is aimed at users of Gemini Enterprise and Workspace services and integrates seamlessly with third-party platforms. The core innovation lies in a shift of approach: instead of providing step-by-step instructions, the user sets objectives for the agent, and the system autonomously determines the execution algorithm.
The functionality of the Gemini agent includes media content creation, writing and running program code, as well as handling professional tasks. For instance, the agent can independently coordinate a meeting by analyzing chat threads and participant calendars without the need for manual oversight.
Model Flexibility
The agent’s technical architecture is designed to select the most suitable model for each specific job. Beyond the Gemini model family, the platform supports operation with Claude, and there are future plans to integrate other private and open-source models to optimize quality and reduce costs.
To tackle multi-step tasks, the system employs sub-agents capable of coordinating workflows in parallel or sequential steps that may run for extended periods. The agent operates in the cloud, maintaining a single set of memories, context, and a personalization graph for the user.
System Memory Layers
Developers have introduced four types of memory to enhance task management. Session memory supports the current task, semantic memory builds a structured knowledge base, procedural memory manages learned skills, and episodic memory tracks all previous activity.
The tool is invoked using the @Gemini command in Gmail, Google Docs, Sheets, Slides, and Chat. Furthermore, a specific API allows for integration into third-party corporate environments, such as Slack and Microsoft 365, operating as a headless agent without the need for a dedicated user interface.
Illustrative photo: Finitehoax / CC0
