Gemini Agent: Google Pitches AI Coworkers for Enterprise Work

What Google announced

Google Cloud has introduced Gemini agent, which it describes as a single "universal agent" for work. It is designed to plan tasks, use tools and skills, and connect to the business systems where corporate work actually happens 1. The announcement came around the company's Gemini at Work 2026 event, where Google Cloud CEO Thomas Kurian presented the product in a keynote 13.

Kurian's framing is the clearest summary of the pitch. Users give Gemini "objectives, not instructions," delegate an outcome, and return later to finished work 1. To do that, he argued, the agent has to be wired into personal workflows, systems of record, and enterprise controls 1.

On capabilities, the coverage broadly agrees. The agent can write and run code, generate images and media, and plug into Google products such as Gmail, Docs, Drive and Sheets 3. It also works with third-party tools including Slack and Microsoft 365 3. Reports cited from TechCrunch and The Verge add that it can delegate to sub-agents and operate across both Workspace and Microsoft 365 2.

Where the accounts diverge

All three outlets describe the same launch, but each stresses something different, and together they give a fuller picture.

Constellation Research focuses on the plumbing 1. Gemini agent routes requests across models, includes cost-control tools, is governed by enterprise policies, and exposes a single API that works from objectives 1. This is the CIO's view of the product: who controls it, what it costs, and how it fits existing governance.

CBS News emphasizes the worker-facing side 3. It positions the agent as a virtual office assistant in a crowded race among tech giants and highlights its coding and media-generation abilities 3.

The Agentic Ready roundup offers the most striking detail. According to the reports it cites, the agent operates with its own email address 2. It also carries Google's scale claims: nearly 90 percent of Fortune 100 companies use Gemini Enterprise, and Gemini has more than 1 billion monthly active users overall 2. It notes the rollout targets business users first, with consumers to follow 2.

The identity question is the real story

Code execution and cross-app integration are now standard for agent products. What stands out here is how Google describes the agent's place in an organization.

Kurian wrote that Gemini can act as a personal assistant or as a team member. It might work for a group, like a project manager, or fill a specific role, like an analyst in a finance department 3. Combine that with the reported dedicated email address 2 and the agent starts to look less like a feature inside an app and more like an entity with a seat in the org chart. Something with an address can receive requests, send work out, and be held accountable in ways a chatbot sidebar cannot.

This also explains why the governance details matter so much 1. If an agent behaves like a coworker, companies need to manage it like one. They have to decide what it can access, whose policies apply, and how much compute it may spend. Model routing and cost controls are Google's answer to the practical worries that come with letting software take on a role rather than respond to a prompt.

Why it matters

The Microsoft 365 support is strategically notable 23. Many large enterprises run on Microsoft's productivity stack. An agent that works only inside Google's ecosystem would have limited reach. By offering cross-platform support, Google can try to win the agent layer even at companies where it doesn't own the documents and inboxes.

The business-first rollout fits the same logic 2. Enterprise buyers pay for governance, integration and accountability. Those are the features that make an identity-bearing agent viable, and they are harder to monetize with consumers.

Caveats

Much of what is known comes from Google's own statements and keynote messaging 13. Usage figures like the Fortune 100 number describe Gemini Enterprise broadly, not adoption of this new agent 2. The promise that users can "come back to finished work" 1 will depend on reliability in messy real-world systems. That is where agent products have historically struggled.

The direction is clear, though. Google is betting that the next phase of workplace AI is about where the AI sits in the company, not just what it can do. Gemini agent is pitched as a colleague with a role, an inbox, and a rulebook. Whether enterprises are ready to staff their teams that way is the question the coming months will answer.