Google Cloud chief Thomas Kurian used the keynote at Gemini at Work 2026 to announce the Gemini agent: a single, universal agent for work that handles questions, knowledge tasks, media creation and coding from one interface. The launch signals that enterprise AI is moving from separate tools for separate jobs to a single system that can work across devices, applications and teams.
The announcement comes with a set of new capabilities including inline access inside Google Workspace, data and analytics skills, industry specialisations for finance and legal, and a framework for governing agents through identity controls, secure sandboxing and cost management.
The numbers behind the shift
Google said nearly 80% of its cloud customers now use its AI products, and close to 90% of the Fortune 100 use Gemini Enterprise. Nearly 500 customers each processed more than one trillion tokens in the past year, suggesting the infrastructure is now operating at a scale that makes single-environment agents practical.
That scale is important because it shows enterprise buyers are not experimenting any longer. They are running production workloads. A universal agent only works when a large enough proportion of a company's data, tools and users sit inside one ecosystem.
What the Gemini agent actually does
The Gemini agent is described as a unified agent that can switch between answering questions, working autonomously on assigned objectives and generating code. Users set an outcome rather than typing step-by-step instructions, and the agent plans the work, selects tools, connects to systems and returns a finished result inside the documents, inboxes and developer environments the user already works in.
Key architectural principles include:
- Omnipresent access. The agent is available on web, iOS, Android, Windows, Mac, command line, Google Workspace, Microsoft 365 and Slack. It can also integrate into third-party applications and operate as a headless agent without its own interface.
- Persistent execution. Work continues in the cloud after the user closes a device. The agent maintains memory and context across devices and sessions.
- Multi-agent orchestration. Gemini can create temporary sub-agents for multi-step tasks, each with its own identity. These sub-agents communicate to coordinate parallel and sequential steps. Longer-running "coworker agents" can have their own email addresses and persistent storage.
- Deep context. The system is already connected to a company's tools and work history. Teams can create dedicated projects to further refine the context and skills available to the agent.
- Model flexibility. The agent chooses the most suitable model for each task, orchestrating across Google's own Gemini models and Anthropic's Claude, with other models expected in the future. Kurian noted that the best model is not always the largest one.
Governance as architecture
Google described its approach to agent governance as built in rather than bolted on. The system uses identity and policy management, authorisation controls, secure sandboxing, network gateways and spending controls including multi-model orchestration, smart routing and real-time spend caps.
For enterprises, this is significant because it addresses three of the biggest barriers to agent deployment at scale: who the agent is and what it can do, what happens if it goes wrong, and how much it costs. Previous deployments across other providers have shown that cost can spiral quickly when an agent is given unlimited access to the most capable models for every task.
Industry signals: from research to workload
Google also introduced industry specialisations for financial services and legal teams, with tools and connectors specific to those sectors. Premium sportswear brand On was named as a business that tested the dynamic model-selection capability.
During the 6 to 9 October 2026 period covered by Gemini at Work, online publication TechCrunch reported that Google was also testing internal models referred to as Gemini 4, described as a frontier model competing with systems from other labs. That line of advancement suggests the commercial agent product and the underlying research are both moving quickly.
What this means for enterprise buyers
For procurement and technology leaders, the Gemini agent represents the arrival of a genuine universal-agent platform inside a major cloud ecosystem. Buyers evaluating agent solutions now face a simpler but more consequential choice: buy a single-agent platform from one of the major providers, or build an orchestrated stack from multiple specialist vendors.
The universal-agent approach reduces integration complexity and gives end users a single interface, but it also increases dependency on one vendor's roadmap, pricing and data-handling terms. Buyers should therefore evaluate two things: whether the agent can access the systems and data they already use, and whether it can be audited, governed and cost-managed under their existing controls.
Google's inclusion of identity management, sandboxing and spend controls in the agent architecture is a signal that it understands those questions. Buyers should still run proof-of-concept deployments focused on governance scenarios before committing to a full rollout.
What this means for suppliers
For companies selling point-solution agents, the Gemini announcement raises the competitive stakes. A general-purpose agent that can plan, generate, research and code from a single prompt window creates direct overlap with specialised tools in knowledge work, content creation and software development.
The advantage for specialist vendors lies in depth: industry-specific workflows, regulated-environment controls, proprietary data or deep integrations that a universal agent cannot replicate without heavy customisation. The task for suppliers is to demonstrate that their solution delivers outcomes the universal agent cannot match at a price justified by that extra capability.
The sub-agent and coworker-agent model also creates partnership opportunities. A company building a specialised financial analyst agent could realistically become a coworker agent that Gemini orchestrates into a broader workflow.
The Agentic Expo angle
The Gemini at Work announcement confirms that the universal enterprise agent is no longer a future concept. It is a product that major cloud customers can already deploy. That shift is exactly why Agentic Expo at Olympia London on 23 to 24 March 2027 matters.
Buyers evaluating agent platforms, security layers and orchestration tools will find suppliers that have already solved the specific challenges universal agents leave unsolved: regulated-sector deployment, deterministic control, audit-quality logging and hybrid architectures that keep sensitive data on premises while connecting to cloud models.