Google Cloud's Gemini agent aims to replace the enterprise AI tool sprawl with one agent and one API

Google Cloud used the opening keynote of Gemini at Work 2026 on October 8 to announce the Gemini agent, which it calls "your new single, universal agent for work." In a keynote write-up attributed to Google Cloud CEO Thomas Kurian, the company describes one agent, reachable through one prompt box and one API, that answers questions, does knowledge work, generates images and media, and writes and runs code.

The pitch is consolidation. Over the past two years, enterprise AI buyers have ended up with a chat assistant, a separate coding tool, a no-code agent builder and a scattering of department-level bots. Google's answer is to fold that into a single agent that plans the task, calls tools and skills, connects to company systems and returns finished work inside the documents, inboxes and developer environments people already use.

What Google says the agent does

According to the keynote post, the Gemini agent runs in the cloud, persists across devices and can carry multi-step work for hours or days. It supports chat, autonomous task execution, scheduled tasks and event-triggered responses, and can spin up temporary sub-agents that work in parallel or in sequence.

• Where it runs: web, iOS, Android, Windows, Mac, the command line, Google Workspace, Microsoft 365, Slack and third-party apps. Google also says it can be embedded in other applications as a headless agent, with no dedicated interface of its own.
• Memory: four types: session, semantic, procedural (including skills the agent writes for itself) and episodic.
• Connectors: Confluence, Microsoft Office and Teams, Slack, Workspace, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres, Snowflake, local files and any MCP server. Enterprises can publish their own tools and skills to internal registries.
• Models: Google says the agent runs each job on the model that fits best, "orchestrating across our Gemini family of models and Claude models from Anthropic today," with other private and open models planned. The post does not say which Claude models are included.

Coworker agents with their own email addresses

The most concrete change for Workspace customers is inline Gemini in Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, plus what Google calls coworker agents. These are agents assigned to a team or a role, such as a finance analyst or an events coordinator, rather than to one person. Google says they have "dedicated identities including their own @agents.company.com emails, their own persistent storage," and only see the context that team members give them. Its example: a marketing manager asks an events-coordinator agent in a group chat to draft a launch-readiness document, and the agent posts it back when it is done.

Giving an agent a mailbox and an identity is a meaningful design choice. It makes agent actions attributable in audit logs, and it also means an agent can send and receive mail like any employee account, which raises the stakes on permissions.

Governance and cost controls

Google spent a large share of the post on control. It lists per-agent identity, least-privilege role-based permissions, OAuth-based identity propagation and audit trails attributed to the agent. Every Gemini agent task runs inside an Agent Sandbox "with its own network boundary," and all traffic in, out and between agents passes through Agent Gateway, which Google describes as an AI network firewall enforcing company policy in real time.

On cost, Google says per-token prices have fallen 98% since 2024 while enterprise volume has exploded, and it is adding to the billing options it announced in August. The new items named today are Smart Routing, which sends each workload to the model that Google says delivers the best performance at the lowest cost, and spend caps: if a project's cap is hit, that project's agent pauses until an admin resumes it with one click in the console.

Industry versions and data skills

Google is also packaging domain variants. The post says Gemini "is now in preview for Financial Services and Legal, and coming soon to Government, Healthcare, and Retail." A data skill set targets data engineers (PySpark generation, notebooks, model training, pipeline fixes) and business users, for whom Gemini can build saved operational reports that, according to Google, run on demand without token costs. That relies on a Knowledge Catalog, in-place enrichment of unstructured data and a lakehouse that queries across S3, Azure Data Lake, Salesforce, SAP, ServiceNow, Workday and Iceberg tables.

What is still missing

Neither the keynote post nor Google's short announcement gives a general-availability date, per-seat or per-task pricing, or a region list for the Gemini agent itself. The only explicit availability label in the keynote is the preview for the Financial Services and Legal versions. The long customer list in the post (BNP Paribas, SOMPO, PayPal, Snap, the U.S. Chief Digital and AI Office and dozens more) mostly describes existing Gemini Enterprise deployments, and the results quoted are company-reported figures, not independent measurements. The same applies to Google's claim that its TPU 8i system delivers 80% better price-performance than the previous generation.

Teams evaluating the Gemini agent should wait for the product documentation before planning around it, and should confirm three things in particular: which Claude models are available and how they are billed, whether coworker-agent mailboxes can be restricted to internal recipients, and how spend caps interact with the existing seat and pay-as-you-go plans.