Google Cloud Launches Gemini Agent for Enterprise Work

The report describes the Gemini agent as a delegation layer rather than a chatbot. Users give it objectives, not instructions, and it plans work, selects skills and tools, connects to company systems, and returns finished output. At Google Cloud's Gemini at Work event, Google Cloud's Thomas Kurian said in an Oct. 8 post that Gemini is a new single universal agent for work with all business context, usable from knowledge work and questions to content creation and coding from a single prompt box. Google outlines six architectural principles: a unified agent interface for chat, autonomous objectives, and code generation; omnipresent access across devices and third-party apps; persistent execution in the cloud with one set of memories and one personalization graph; multi-agent orchestration through temporary sub-agents with their own identities; deep context learned over time; and flexibility to run each job on the best-fit model. For developers, the report says, the agent is the product and the model is a routing decision.

Memory and reasoning are built around four memory types. Session memory covers the current task, even across days. Semantic memory is a structured knowledge base built from documents and people. Procedural memory stores how jobs are done, including skills the agent writes for itself. Episodic memory records what it has done before. Skills are modular prompts stored in a shared company registry, tools come from an enterprise tools registry, and connectors cover Slack, Jira, Salesforce, ServiceNow, BigQuery, Snowflake, desktop files, and any Model Context Protocol server. The agent currently orchestrates across Google's Gemini models and Anthropic's Claude models, with other private and open models planned. Google's own lineup includes Argon for frontier reasoning, Flash for speed and volume, Omni for generative media, and Gemma for open-weights edge work.

A coworker agent is a persistent teammate with a defined role. In Workspace it receives its own account, including email address, calendar, Drive, and a directory entry. Colleagues can @mention it in Chat or Docs, and its edits appear under its own name in version history. It sees only what is shared with it, works inline across Gmail, Docs, Sheets, Slides, Chat, and Calendar, and offers one-click delegation of tasks it spots. For data teams, data and machine learning engineers describe outcomes in plain language, and the agent writes PySpark code, provides notebooks, trains models, and fixes pipeline issues. Business users get saved BigQuery reports that rerun without token costs.

Three services ground answers. Knowledge Catalog maps business definitions once for every agent. Smart Storage enriches unstructured objects in place; Google says 90% of enterprise data is unstructured. Borderless Lakehouse queries Amazon S3 and Azure Data Lake with no variable egress fees. MarkTechPost cited Bloomberg Media as lifting SQL query accuracy by 63% by grounding data agents in Knowledge Catalog.

Governance is framed around four questions: who, what the agent may do, what it did, and what it must never touch. Each agent gets a cryptographically attested identity with least-privilege permissions. Authorization uses role-based access mapped to external systems through standards such as OAuth. Every action is logged to the agent rather than a person. Agents run in an Agent Sandbox, and all traffic passes through Agent Gateway, described as an AI network firewall that applies one policy to every agent.

Cost controls rely on multi-model orchestration, Smart Routing that triages workloads to the cheapest capable model, and real-time spend caps. Teams can set a hard project limit in the Cloud Billing Console; when triggered, that project's agent pauses until someone resumes it. Google says its TPU 8i system delivers 80% better price-performance than the prior generation. The report says buyers must evaluate the product partly on customer anecdotes rather than reproducible numbers, and it notes that the comparison with Microsoft 365 Copilot, OpenAI ChatGPT Work, and Amazon Quick was not fully detailed in the available material.