Last updated: 10 October 2026, 10.38pm BST.
Google Cloud has introduced a new Gemini agent designed to handle work across documents, email, code and business systems from one interface, pushing its enterprise AI offer beyond chat and into longer-running delegated tasks.
What Google says the Gemini agent can do
In the original Google Cloud announcement, chief executive Thomas Kurian describes Gemini as a single work agent that accepts an objective, plans the job, selects skills and tools, and returns completed work.
The agent is intended to operate across web, mobile, desktop and command-line environments. Google says it can work inside Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, and can connect to third-party systems including Microsoft 365, Slack, Confluence, Salesforce, ServiceNow, Git, Jira and databases.
Google also says cloud execution lets a task continue after a laptop is closed. Memory is divided into session, semantic, procedural and episodic forms, while multi-agent orchestration can create temporary specialist agents for parts of a larger job. Organisations can create longer-lived “coworker” agents with defined identities and restricted access.
What is actually new
The important change is consolidation. Rather than presenting separate chat, coding and automation products, Google is trying to make one agent the starting point for work and let the underlying model vary by task. The announcement says Gemini can route work across Google models and Anthropic’s Claude models, with more options planned.
That model flexibility is potentially useful for cost and quality, but the launch material does not provide a neutral comparison showing when routing improves an outcome. Businesses should treat performance percentages and customer case studies as vendor evidence until independently reproduced.
Security and cost questions
Google highlights identity controls, permissions, sandboxing, policy management and real-time spending limits. These controls matter because an agent connected to email, files and business systems can cause more harm than a chatbot if permissions are too broad or a task is misunderstood.
Organisations considering the product should ask five practical questions: what data is retained; which connectors can write as well as read; how approvals work for external actions; what audit logs administrators receive; and how costs behave during long tasks or sub-agent activity.
Google publishes separate Agent Platform pricing, but product configuration and model choice can change the final bill. A token price alone is not a reliable estimate of the cost of an autonomous workflow.
Why it matters beyond Google
The launch shows where enterprise AI competition is moving: from answering prompts to operating across tools, retaining context and completing work over time. Sussex’s own technology sector is active too; our report on Universal Quantum’s latest funding covers a very different part of the regional innovation economy.
Brighton Locals’ view
Our assessment: a single agent that can work across an organisation is more coherent than a collection of disconnected assistants. It is also a larger governance risk. The most persuasive test will not be a polished demo, but whether companies can limit permissions, reproduce decisions, control spend and recover cleanly when an agent is wrong.
What to watch next
Watch for a clear availability timetable, UK pricing, independent benchmarks and documented failure rates. Buyers should pilot the system on reversible, low-risk work before giving it authority over customer communication, payments, production code or regulated data.
Bottom line: Google has set out an ambitious universal work agent. The feature list is substantial, but trust will depend on evidence from real deployments rather than launch-day claims.
Featured image: Google Gemini icon. Google LLC / Wikimedia Commons. The simple geometric logo is marked public domain for copyright purposes; Google’s trademark rights remain.
Leave a Reply