Google’s Gemini Agent Puts Workplace Delegation to the Test

Google’s announcement of a new Gemini agent in Gemini Enterprise combines conversational assistance with autonomous task execution. Alongside it, Google is introducing AI coworkers, persistent agents with their own workplace identities, permissions and activity records.

The ambition is to take on more of the coordination surrounding everyday work, reducing time spent finding information, following up and moving tasks between applications.

Employees’ experience of AI assistance doesn’t automatically translate into readiness to delegate responsibility. Data from FDM CCS Insight’s forthcoming Survey: Employee Workplace Technology, 2026 shows that 88% of generative AI users report faster task completion. Yet over two-thirds of all employees would currently limit an agent to preparing material for review, recommending actions or completing tasks with approval before every action.

Google is bringing together capabilities that could make delegation more practical; the value will depend on how much work employees can confidently hand over and how consistently agents deliver the intended result.

Reducing the Work between Applications

The agent is designed to work on browsers, mobile and desktop apps, and workplace tools including Google Workspace, Microsoft 365 and Slack. Running in the cloud allows it to retain context and continue working in the background as employees move between tools.

Within Workspace, the agent works directly in Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. Google gives the example of arranging a meeting by identifying participants, checking calendars and coordinating attendees.

The opportunity is to reduce some of the digital friction that employees experience as they move work between systems. Knowledge work still relies on people carrying context between applications: finding the right document, copying information into another tool, checking who needs to be involved and following up when something stalls. An agent that can maintain context across those boundaries could remove some of that coordination burden rather than simply making individual tasks faster.

Our survey also finds that 58% say collaboration and productivity applications from the same supplier work together very well, compared with 26% for applications from different suppliers. The depth of integration therefore matters, particularly where completing a task requires actions spanning several suppliers’ systems.

Google is also introducing proactive delegation. Workspace Intelligence can recognize a task in an incoming message — such as a manager asking for a project update as a slide deck — and offer the employee a one-click option to hand that work to Gemini. This could make delegation a more-natural part of daily work.

Teams can also tailor the context, skills and tools the agent uses. FDM CCS Insight’s report, Scaling Agentic AI: Foundations, Partners and Next Steps, highlights why that grounding matters: persistent memory can’t compensate for outdated records, conflicting instructions or unclear ownership.

Giving AI Coworkers Clear Boundaries

Google’s coworker agents bring a more-lasting role for AI into the team. Employees describe the role they need and the agent receives its own Workspace identity, allowing colleagues to involve it in work while keeping its contributions attributable.

Giving the agent a visible workplace identity is important. As AI takes on more work, employees need to be able to see when they are interacting with a person, an agent or work produced by an agent. A separate account and attributable actions can make delegation easier to understand and challenge, while making responsibility more explicit than automation running invisibly in the background.

For more-complex work, temporary agents can take on different parts of a task, making oversight increasingly important as responsibility passes between them.

Each coworker agent reports to the person who created it; this person is responsible for managing its permissions and can transfer responsibility to someone else. This provides a starting point for ownership. Organizations still need to establish who approves actions, oversees the agent’s role and is ultimately accountable when something goes wrong.

FDM CCS Insight’s Survey: Senior Leadership IT Investment, 2026 reinforces these requirements. Among leaders whose organizations are exploring, piloting or using agentic AI, almost two-thirds say accountability for failure matters more than headline claims of autonomy. Meanwhile, 70% believe human approval should remain mandatory for higher-risk actions.

Google’s safeguards include separate agent identities, restricted access, administrator-approved permissions and records of agent actions, alongside controls for task execution and agent traffic.

Bringing these controls into the same offering as agent execution is significant because organizations need to manage not only what an agent can access, but what it’s allowed to do. In our report, Agentic AI: The Supplier Reality, we stress that permissions, policy enforcement and oversight must apply to actions as well as systems. Buyers should check that these controls remain effective when agents delegate work or use external applications, and understand where additional safeguards or configuration are still required.

Recovery also needs scrutiny, as the ability to correct mistakes will shape how much responsibility organizations can confidently delegate. Restoring a document may be straightforward; reversing an external email or an action from a business system may not be. Organizations need to decide where approval is required before an agent acts, alongside how they’ll stop activity, correct errors and manage any wider impact.

Google is also separating the agent from its underlying model, allowing organizations to use different models for different tasks. That flexibility is useful, but changing the model can alter how an agent interprets instructions and completes work. Organizations will therefore need to keep testing behaviour as models, prompts and tools change, while defining clear guardrails that apply regardless of the underlying model.

The Test Is How Delegation Works in Practice

Google’s announcements bring personal assistance, team agents and coordinated execution together with controls over their actions. Working inside familiar applications could make these capabilities easier to incorporate into daily routines, and persistent context could reduce the effort of keeping work moving.

Employers will need to prepare people to define objectives, set boundaries and assess results. Giving an agent a role in a team also creates responsibilities for its owner and for colleagues who assign it work, making clear guidance essential.

For Google, the next test is evidence that the agents reliably complete tasks across connected systems, with employees able to understand, intervene in and recover from the actions that agents take. Organizations should begin with defined workflows, measure the total effort involved and expand usage where results justify it. The value of an AI coworker will depend on whether employees can entrust it with useful work and spend less time managing that work themselves.