Google Introduces Gemini AI Agent Capable of Working on Behalf of Office Teams
Google has announced a new Gemini AI agent designed to take on workplace tasks not only for individual employees but also on behalf of groups, teams, or specific roles within an organisation. The announcement was made during the Gemini at Work 2026 event, where Google positioned the agent as a single AI system capable of answering questions, handling knowledge work, creating media, and writing and running code.
According to a blog post by Google, the Gemini agent can function as a team member, such as a project manager within a team, or represent a specific role, such as an analyst in a finance department. Users can provide the agent with an objective rather than explaining every individual step required to complete a task.
Coordinating Complex Workplace Tasks
The Gemini agent is capable of managing tasks that typically require coordination among multiple people. For example, if a manager requests a slide deck with the latest project update via email, Gemini can recognise the task and offer a one-click option to complete it.
The agent can also coordinate meetings. An employee can instruct Gemini to arrange a meeting with regional event leads without providing specific names or email addresses. The agent can identify the correct individuals by analysing chat space memberships and previous conversations, check their calendars, and initiate an email thread to find a suitable time, even when external participants are involved.
Additionally, teams can establish "coworker agents" with defined roles, unique identities, company email addresses, and persistent storage. These coworker agents only access information provided by the user or team. Gemini can also generate temporary sub-agents to handle different components of complex assignments.
Integration and Cloud Functionality
Because the agent runs in the cloud, tasks that require hours or days to complete can continue running after an employee closes their laptop. The system is accessible across web, Android, iOS, Windows, and Mac platforms, as well as workplace services including Google Workspace, Microsoft 365, and Slack.
Gemini can connect with various business tools and databases, such as Salesforce, ServiceNow, Jira, Git, BigQuery, Databricks, Postgres, and Snowflake. Organisations also have the option to develop their own custom tools and reusable skills for the agent to utilise.
Anthropic Study Highlights Uneven AI Exposure
While Google expands the capabilities of its workplace AI, research from Anthropic indicates that the impact of AI remains uneven across different professions. The study analysed the real-world usage of Anthropic's Claude AI and compared it with theoretical AI capabilities, measuring "observed exposure" to determine how much of a job is actually performed using AI.
Anthropic found that several professions maintain low AI exposure due to requirements for physical activity, real-world interaction, or responsibilities that current AI systems cannot perform. These include roles in agriculture, construction, installation and repair, transportation, personal care, and food services.
For instance, agricultural workers must prune trees or operate machinery, and legal professionals representing clients in court must be physically present, limiting the involvement of language models. Conversely, digital occupations show much higher AI involvement, with computer programmers having about 75 per cent of their tasks showing AI involvement. Customer service representatives and data entry keyers also rank among the more exposed roles.
The research also identified a gap between theoretical potential and actual usage. In computer and mathematics occupations, AI could theoretically assist with nearly 94 per cent of tasks, but actual Claude usage currently covers only about 33 per cent.
