The prompt becomes a work order
Today, October 8, Google Cloud unveiled the Gemini agent at its Gemini at Work 2026 event, a single layer meant to replace the old question and answer chat with a system that takes objectives and returns finished work. The company sums up the shift in one line: you delegate an outcome and come back to find the task done. The agent answers questions, handles knowledge work, creates images and other media, and writes and runs code, all from a single prompt box. It reaches users through email and documents in Google Workspace, through Microsoft 365, through Slack, through the command line, and through an application programming interface that lets any third party app call it as a headless agent with no screen of its own.
Wired into everything a company already runs
In practice, the Gemini agent connects to collaboration tools such as Confluence, Teams and Workspace, to development tools such as Git and Jira, to enterprise platforms such as Salesforce and ServiceNow, to databases such as BigQuery, Databricks, Postgres and Snowflake, to files on a user's desktop, and to any server that follows the Model Context Protocol, an open standard for linking agents to outside systems. It also ships with a catalog of reusable skills that teams can publish company wide, and it learns from every execution to repeat the cheapest, most accurate path next time.
The detail that stings anyone selling exclusivity: the engine is also Claude
The most revealing choice in the announcement is not the interface, it is the engine underneath it. The Gemini agent routes each task to whichever model fits best, and today that already includes both the Gemini family and the Claude family from rival Anthropic, with a promise to add private and open models later. Google itself is admitting, in bold print on its own product launch, that betting only on in house technology stopped being a winning strategy. For anyone choosing an automation tool at a small or medium business, the practical takeaway is direct: multi model architecture, already common in third party orchestrators, is becoming factory default even inside the largest cloud infrastructure competitor. Brand loyalty matters less and less than the result per task.
It keeps working after you close the laptop
The agent runs in the cloud and keeps a single set of memory across devices and sessions, split into four layers: session memory for the task at hand, semantic memory built from documents and conversations, procedural memory for how each job gets done, and episodic memory of everything it has already executed. That lets a task spanning hours or days keep running after the user steps away from the computer. The agent also spins up temporary sub-agents for specific steps and can take on a fixed role inside a team, called a coworker agent, with its own email address on the company domain, dedicated storage, and access limited to only what it was given.
Security comes bundled, but the numbers are Google's own
Every execution runs inside an isolated sandbox, and all traffic passes through a network firewall built specifically for agents, the Agent Gateway, which applies a rule written once to the company's entire agent fleet. Google leans on customers as social proof: Orange Spain already runs more than a thousand custom agents, Japanese insurer Sompo reached ten thousand agents across 34,000 employees, and Bradesco reports document review time dropping from one hour to five minutes, with a 60 percent cut in risk inconsistencies. These are figures supplied by the company itself, not an independent audit, but they signal where the enterprise market is heading.
What is still missing before this reaches daily use
The Gemini agent is in private preview, and Google promises broader availability soon only for Workspace accounts on select Business and Enterprise plans. There is no public pricing, no general launch date, and no list of supported countries. MaxAssistant's take: the direction is unmistakable, chat is giving way to an agent that takes an objective and returns a delivery, and that applies to Brazilian marketing, support and automation operations that today juggle several separate tools. But migrating before pricing and regional availability are confirmed means betting on a stage promise, not a testable product. Worth watching closely and testing the moment the account opens.
Sources
Google: Google Cloud introduces the Gemini agent; Google Cloud: Welcome to Gemini at Work 2026; 9to5Google: Google Cloud announces Gemini agent.