Google's New Gemini AI Agent Gets Its Own Email Address and Spending Limit

Enterprise AI Agents
Google's new Gemini agent can get its own mailbox, and your administrators set its spending limit. Here is what your team can hand off, who competes, and what to settle before you do.
By Shashi Bellamkonda · October 9, 2026
~90%
of the Fortune 100 use Gemini Enterprise (Kurian)
63%
query accuracy gain at Bloomberg Media in early development (Kurian)
98%
drop in per-token prices since 2024 (Kurian)
$360
Microsoft 365 Copilot list price per user per year (O365HQ)

Its own email address is the detail in the Google Cloud Gemini agent announcement that stopped me. I wanted one for my own AI assistant so I could forward it email, and it has no inbox to forward to. Google gave its agent one.

Google announced the Gemini agent on October 8 at Gemini at Work 2026 as a single assistant for work. You describe an outcome, such as a slide deck on last quarter's sales, and it plans the steps, pulls data from your systems, and returns the result in the tool where you work, whether that is Gmail, Docs, or Slack. It keeps one memory across Google Workspace, Microsoft 365, a command line, and your phone, and it picks the artificial intelligence model for each task from Google's Gemini family and from Claude models built by Anthropic.

A coworker agent gets its own mailbox and appears under its own name in version history

You describe the role you need, and Gemini creates the agent. It receives its own Workspace account with an email address, a calendar, a Drive, and a place in your company directory. Colleagues add it to a Chat space or mention it by name, the way they would reach any teammate.

Google's example is a marketing manager who asks an events coordinator agent in a chat group to draft a launch readiness document, and the agent posts it back to the group when it finishes. The manager can tag the agent in a document comment, where it suggests an edit and replies in the thread, and its changes appear under its own name in version history. A coworker agent acts under its own identity and sees only what you share with it.

Three other patterns from the keynote:

  • Scheduling without names. Ask it to set up a meeting with your usual regional event leads, and it works out who they are from your chat space and your last event thread, checks their calendars, and starts an email thread to find a time.
  • Handing off from your inbox. When your manager emails asking for a project update as a slide deck, Workspace offers a one-click option to pass the job to the agent.
  • Data questions in everyday words. Ask for an operating report, and the agent writes the query, runs it in BigQuery, and saves it so teams rerun the report on demand without paying token costs again.

Work that takes hours or days keeps running after you close your laptop.

Early gains show up in data work, where Bloomberg Media raised query accuracy 63%

Bloomberg Media grounded its data agents in Google's Knowledge Catalog, which maps business terms such as net margin once so every agent uses the same definition. During initial development its structured query language accuracy rose 63% (Kurian). Deloitte says an agentic blueprint built on Gemini gets clients to measurable business impact up to three times faster (Kurian).

Both figures come from Google's keynote and describe early work, so ask each vendor contact what baseline the gain was measured against. Industry versions are further out. Gemini for Financial Services and Legal is in preview, with CME Group and Deutsche Bank named as users of the financial services version, and Government, Healthcare, and Retail are listed as coming soon.

Four rivals sell the same promise, and each one prices it in a different unit

  • Microsoft 365 Copilot. Lists at $360 per user per year, and a seat includes agents built in Copilot Studio with no metering for that user. Pay-as-you-go agent use costs $0.01 per message, and O365HQ reports that Microsoft's budget feature sends alerts without capping usage.
  • Amazon Quick. Launched as Quick Suite on October 9, 2025, with a Professional tier at $20 per user per month and an Enterprise tier at $40, each including agent hours (YourStory). Amazon Web Services may have changed these prices since launch.
  • OpenAI workspace agents. Announced April 22, 2026 for ChatGPT Business, Enterprise, and Edu plans. They run on schedules, work inside Slack, pause for approval before sensitive actions, and administrators can suspend them. Credit-based pricing began after a free period ended May 6, 2026.
  • Salesforce Agentforce. Charges $500 per 100,000 Flex Credits, and a standard action costs 20 credits, or $0.10 (Salesforce). It works strongest inside Salesforce's own customer records.

Each rival publishes a billing unit: a seat, a message, a credit, an action. Google's announcement gives you a hard spending limit per project, where an administrator sets the cap and the agent pauses when it triggers, but no unit price to set the cap against. Google's pause differs from the budget alerts O365HQ describes for Microsoft.

The audit trail names the agent, so the person who delegated the work has to be named somewhere else.

Five roles in your company have a decision to make on this announcement

  • Workspace administrators. Coworker agents appear in your company directory with their own accounts, so you decide who may create one.
  • Security leaders. Each agent gets its own identity, every action is logged to the agent, and Agent Gateway applies one policy to every agent, such as "agents may not open documents classified Need to Know."
  • Finance and cloud cost owners. Spend caps and department chargeback live in the Cloud Billing Console, per project.
  • Data leaders. Google says the agent reads from Salesforce, SAP, ServiceNow, and Workday without copying data.
  • Chief information officers and chief technology officers. Committing to the agent means committing to its memory, its skill registry, and its gateway.

The announcement does not say when general availability begins or what the agent costs.

The time it saves comes from context that follows you across apps and devices

A single memory means you stop re-briefing an assistant each time you switch from your laptop to your phone or from Docs to Slack. Reading data where it sits saves the copying step: Google says its Lakehouse queries Amazon S3 and Azure Data Lake with no variable egress fees. A saved report reruns without token costs, and one written policy replaces a check on each agent, one at a time.

Some of the effort moves to your administrators. Every coworker agent needs a named human owner on the day you create it. Decide what mail an agent may read and who reviews what it sends, because the announcement describes the mailbox and leaves those rules to you. Every project needs a spend cap, and without a per-task price from Google, your first cap is a guess.

CIO/CTO Viability Question
Before you give an agent a mailbox, ask Google Cloud to put two answers in writing: the price of a typical agent task, and the export path for your skills and memory if you leave. If either answer is a roadmap slide, how much of your work are you willing to build on the agent before the answer arrives?
Sources

Kurian, Thomas. "Welcome to Gemini at Work 2026: Introducing the Gemini Agent." Transformation Today, LinkedIn, 8 Oct. 2026, linkedin.com.

O365HQ. "Microsoft 365 Copilot Licensing and Pricing 2026." O365HQ, 14 Sept. 2026, o365hq.com.

OpenAI. "Introducing Workspace Agents in ChatGPT." OpenAI, 22 Apr. 2026, openai.com.

Salesforce. "Agentforce Pricing." Salesforce, salesforce.com.

YourStory. "AWS Launches Quick Suite, an Agentic AI Workspace." YourStory, Oct. 2025, yourstory.com.

Disclaimer: This blog reflects my personal views only. Content does not represent the views of my employer, Info-Tech Research Group. AI tools may have been used for brevity, structure, or research support. Please independently verify any information before relying on it.