Google Cloud introduced the Gemini agent at a Thursday event in New York, a single AI agent for work that can answer questions and perform tasks on a user’s behalf through one interface, with businesses first in line for rollout. The launch follows consumer-facing agents from competitors including Meta and ChatGPT, and is the clearest sign yet that Google sees agentic AI, not chat, as the next battleground for enterprise software.
The move was covered by TechCrunch, which reported live from the event. Google CEO Sundar Pichai opened the presentation by noting that Gemini already counts more than 1 billion monthly active users, and that nearly 90% of Fortune 100 companies now use Gemini Enterprise at work.
Objectives, not just instructions
Google Cloud CEO Thomas Kurian said the new agent can be given “objectives, not just instructions,” meaning it plans its own work, uses custom skills and tools, and connects to a company’s internal systems to carry a task through. A user can hand over a project brief with attached files and the agent decides how to divide it up, delegating parts to subagents along the way.
By default the agent picks its own model for a given task, but users can override that and choose a specific one, including third-party models, starting with Anthropic’s Claude. Google plans to expand the picker to open source and other private models later.
The agent connects to Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, Snowflake and other business tools. It also works with any Model Context Protocol server, inside or outside a company’s network, which is a nod to the MCP standard that several other AI vendors have adopted this year.
A coworker with its own email
The most notable design choice: the agent gets its own Workspace account, like another employee. It has its own email address, its own context about the company, knowledge of who is on what team, their time zones, who signs off on what and what sits on people’s calendars. Workers reach it by tagging, emailing, sharing, or adding it to a group chat.
As it works, the agent writes an audit trail attributed to the agent itself rather than to any person. Users track progress in a tasks inbox that shows what Gemini is thinking, which subagents it has delegated to, what skills it loaded and any code it produced. That audit structure is meant to answer the compliance questions enterprise buyers ask before granting write access to internal systems.
The agent is reachable from iOS and Android, Windows and Mac desktops, the command line, Google Workspace, Microsoft 365, ServiceNow and Slack.
| System | Availability |
|---|---|
| Google Workspace | Yes |
| Microsoft 365 | Yes |
| Slack | Yes |
| ServiceNow | Yes |
| iOS and Android | Yes |
| Windows and Mac desktops | Yes |
| Command line | Yes |
Early users and pricing
Early testers named by Google include On, Shopify and PayPal. Enterprise customers already using Gemini features include BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty and Wesfarmers.
Flexible spending options are part of the launch. Multi-model orchestration, smart routing, and real-time spend caps let companies assign cheaper models to lower-priority tasks and set a hard cap on what a highly capable model can consume. For organizations that budget AI costs across hundreds of teams, this is practical infrastructure, not marketing.
The agentic race
The competitive map has filled in quickly. Meta launched Muse as a consumer-facing agent. ChatGPT released its own agentic tool under the Dots name. Instinct and other companies have pushed agents that operate over messaging apps rather than websites. Google’s bet is different: rather than building a consumer-first assistant and hoping enterprises follow, it is starting with the large share of Fortune 100 companies already using its workspace products.
Pichai called this approach a way to solve “harder problems around security, scale, and performance” that come with deploying powerful agents, before expanding to consumers later. Google has not said when the general rollout reaches the consumer side of Gemini.
What the launch does settle is the product direction. For two years, Google has been accused of shipping chatbot features while rivals raced to agents. Today’s announcement, and the infrastructure sitting behind it in Gemini Enterprise, shows the company treating agentic work as the main product, not a bolt-on. The billion-user distribution advantage it holds in consumer Gemini gives it an option most rivals lack, but the enterprise pitch is what the market will judge now.
Analysts are split on the shape
Opinion differs on whether enterprises want a single agent touching every system, or a collection of narrow ones tied to specific tools. The tasks inbox and audit trail design suggest Google has heard the concern and is trying to keep the agent inspectable rather than opaque.
There is a deeper question underneath. If an agent writes to production systems, financial records or customer files, who signs off when something goes wrong? Google’s answer so far is attribution: the audit trail names the agent, the human who issued the objective, and the subagents involved. That works for most compliance regimes, but several security teams that Google briefed in recent months, per reporting around the launch, have pushed for tighter scope controls before deployment at scale.
Human oversight is still built in. The agent asks for approval before irreversible actions, and an admin policy layer in Gemini Enterprise can restrict what tools and data any given agent can reach. Those guardrails will decide whether the product lands as a real coworker or a very expensive demo.
For now, the企业 software market has a new reference point. Three consumer agents and one enterprise-first one now sit in the market at the same time, and the battle for the next wave of productivity spend just got sharply more direct.