Google's new Gemini Enterprise Agent tab moves beyond single prompts to run multi-step work tasks on its own, rivaling Anthropic and OpenAI.
Google is testing a new way to use Gemini at work -- not as a chatbot you prompt, but as a system that pursues goals on your behalf. TestingCatalog, which monitors unreleased Google features, has spotted an "Agent" tab inside Gemini Enterprise that doesn't exist in the current public build.
The tab introduces a structured dashboard with panels for Goals, Agents, Connected apps, and Files. A dedicated Tasks inbox sits alongside the usual chat interface. The design is built for multi-step workflows: the user sets a broad objective, and the system breaks it down, routes subtasks to specialized agents, and pulls in the relevant files and services without needing a prompt for every step. Android Authority first reported the screenshots.
One detail stands out in the early build: a "Require a human review" toggle. Before an agent acts -- sending an email, touching a document, interacting with a connected service -- a human must approve. That is a deliberate guardrail in an industry still debating how much autonomy AI systems should hold over live accounts and real data.
The competitive frame
The timing is not accidental. Enterprise AI has become the main battlefield. Anthropic already offers agentic capabilities targeting business workflows, and Google's new interface puts Gemini squarely in the same lane. The broader industry has converged on a shared thesis: the next revenue category isn't chatbots answering questions, it's agents completing tasks.
OpenAI is running the same race with urgency sharpened by IPO pressure. In a March all-hands meeting, Applications CEO Fidji Simo told employees the company is "orienting aggressively" toward high-productivity use cases. The goal, per CNBC, is converting 900 million weekly ChatGPT users into "high-compute users" ahead of a potential Q4 public offering. To support that push, OpenAI plans to nearly double its headcount from 4,500 to 8,000 by year-end, with most new hires going into engineering, product, and a new "technical ambassadorship" role aimed at enterprise adoption, according to Yahoo Finance.
Chrome gets the same treatment
Google is layering agentic logic into Chrome at the same time. A Skills Library, reported by MacRumors, lets individual users save custom Gemini prompts as reusable Skills, triggered by a forward slash or a single click. Build a prompt to analyze a product's ingredient list while shopping -- save it once, reuse it on any page. Google ships the feature with pre-written Skills for common tasks and requires confirmation before any action touching calendar events or email.
The Chrome Skills Library and the Enterprise Agent tab serve different users: one for individual browser sessions, one for team-level workflows. Together they trace a coherent strategy. Gemini becomes a persistent layer across Google's surface area that takes action on your behalf, not just generates text for you to act on.
What this shifts
The move from prompt-response to goal-execution changes the risk calculus in ways that the interface alone doesn't resolve. When an agent operates inside a company's connected apps -- reading documents, querying systems, sending messages -- questions of data access, audit trails, and error accountability become live. Google's human-review toggle is an interface-level answer; what the underlying permissions and logging look like at enterprise scale is not yet disclosed.
Enterprise buyers have absorbed this pitch before. RPA vendors spent years selling software robots that would automate knowledge work. Many deployments turned brittle and costly to maintain. AI agents are more adaptive, but the organizational questions -- who owns the agent, who is liable for its mistakes, how do you audit its decisions -- remain largely unsolved. Google and its competitors are selling the capability before the governance frameworks exist.
The next milestone to watch is Google I/O in May. If Gemini agents appear on stage there, the gap between the leaked interface and a shipping product is probably short. Which Workspace tier gets access first, and at what price, will matter as much as the feature itself.
---
FAQ
What is the new Agent tab in Gemini Enterprise?
It is an unreleased interface spotted by TestingCatalog that lets users create and deploy multiple AI agents for different workflows, with panels for goals, connected apps, and files -- all managed from a single dashboard.
How does Google's agentic AI differ from a standard chatbot?
A chatbot responds to one prompt at a time. An agent takes a broad goal, breaks it into steps, uses connected apps and files autonomously, and executes across multiple actions without needing a prompt at each stage.
Does Google require human approval before agents act?
Based on the early build, yes. A "Require a human review" toggle is visible in the interface, meaning agents are expected to pause and confirm before taking consequential actions.
When could Gemini agents become publicly available?
No release date has been announced. Google I/O in May is the most likely venue for an official reveal, though rollout timing and which Workspace tiers get access first remain unknown.