Google's Gemini Agent: What an AI Coworker With Its Own Email Means for Your Business

Google announced an AI agent that gets its own email, calendar and audit trail. Here is what it does, what it costs, what is still unknown and how to prepare.

13 October 2026 · Deixtra

On 8 October 2026, Google Cloud announced an AI agent that gets its own email address. You describe a job, say "events coordinator" or "accounts follow-up", and Google creates an agent with a Workspace account, a calendar, Drive storage and a line in your company directory. Colleagues can email it, tag it in a chat or add it to a meeting, and it keeps working after everyone has gone home.

Google Cloud CEO Thomas Kurian summed up the pitch at the Gemini at Work event: "Work now starts in the prompt window."

We build business software and AI agents for companies in Pakistan and the Gulf, and we read announcements like this with one question in mind: what can a 20-person company actually do with it, and when? This post covers what Google announced, what's still unknown, and what we'd do about it now. We haven't used the product ourselves, because it's in limited preview. Everything here comes from the coverage linked at the end.

What is the Google Gemini agent?

The Gemini agent is an AI agent for work that Google Cloud announced on 8 October 2026. You give it a goal and it plans and carries out the steps across Gmail, Docs, Sheets, Slides, Drive, Chat and Calendar, and in Microsoft 365 and Slack. It runs in Google's cloud, so a task can continue for hours or days without your laptop being open.

Google calls it a "universal" agent because it combines three things that used to be separate products: a chat assistant, a background worker that completes tasks on its own, and a coding agent.

The examples Google gave are ordinary office jobs. Schedule a meeting with a regional team after checking everyone's calendar. Research a market, build a financial model in Sheets, then turn both into a slide deck. For large jobs it creates temporary helper agents, and each helper has its own identity in the audit log.

What is a coworker agent?

A coworker agent is a permanent agent with a standing role on a team. A manager describes the role, and Gemini sets it up with its own Gmail address, calendar, file storage and directory listing. It shows up under its own name in a document's edit history, and it can only see what the team shares with it.

That last part is the real change. Until now, most AI assistants have acted as you, with your login and your access to everything. If one of them sent a wrong email or deleted a file, the log said you did it. A coworker agent acts as itself. Reports say each agent gets a cryptographically attested identity, and every action is recorded under the agent's name in an audit trail that's separate from any person's.

For a business owner, that makes an AI agent something you can manage the way you manage a new hire: give it access to the sales folder but not payroll, and check what it did last week.

What can it connect to?

According to Reworked's report, the agent ships with connections to Slack, Microsoft Teams and Office, Google Workspace, Confluence, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres and Snowflake, plus files on an employee's desktop.

For anything else, it supports the Model Context Protocol (MCP), an open standard for connecting AI agents to software. This is the part that matters if you run a custom ERP, a POS or an inventory system. If that software has an MCP connector, or someone builds one, the agent can read from it and act in it.

Which AI models does it use?

Google's own Gemini models and Anthropic's Claude models, with others promised later. You can pick a model, or let the agent choose one for each job based on quality and cost. Google says a company can change models later without losing its data or settings.

It's unusual for Google to route work to a competitor's model inside its own product. One line from the coverage explains the thinking: if Google can't always have the best model, it can still be the place where the work gets done.

How much does the Gemini agent cost, and when can you get it?

This is where the announcement is thin. Here's what's been reported:

Question	What's known
Is it available now?	Only to a limited number of companies, in preview
When is general release?	Google hasn't given a date. It says "soon" for Workspace customers on select Business and Enterprise plans
What does it cost?	A Google spokesperson said it's available "at no additional charge for customers where Gemini Enterprise is available". Google hasn't said whether every feature, including coworker agents, is covered
Are there industry versions?	Financial services and legal are in preview. Government, healthcare and retail are listed as coming next
Is it available in Pakistan or the UAE?	We found no country list

"No additional charge" needs a careful read. Agents use tokens, and tokens cost money. Google built spending controls into the product for a reason: admins can set a hard spending cap per project, and when the cap is reached the agent pauses until someone clicks resume. Spending is tracked per project so it can be billed back to a department. Google itself acknowledged that AI budgets are under strain as usage grows, even though the price per token keeps falling.

If you want a feel for what agent usage costs in practice, we worked through the numbers in how much an AI agent costs in 2026.

How does Google keep the agents under control?

Google framed it as four questions a company should be able to answer about any agent: who is it, what is it allowed to do, what did it do, and what should it never touch?

The controls reported so far are:

Each agent has its own identity and role-based access, limited to what it has been given.
Every action is logged under the agent's name.
Agents run inside what Google calls an Agent Sandbox.
Their traffic passes through an Agent Gateway, which Google describes as an "AI network firewall". A rule written once applies to every agent.
When an agent connects to other software, it signs in with the standard "Sign in with Google" system.

Analyst Carmi Levy told Computerworld that Google is trying to become "the gatekeeper of the new enterprise operating system", and that the controls are the main selling point. The same coverage notes that the real test comes when businesses use it widely in production. That's fair. None of this has been through a year of real companies making real mistakes with it.

What does this mean for a small or mid-sized business?

Three things, in our view.

First, AI agents are becoming a standard feature of office software. If your company already pays for Google Workspace, an agent will probably appear in it without a separate purchase. You won't need to go looking for one.

Second, the hard part moves from "can AI do this?" to "what do we let it touch?". An agent that can read your shared drive will read all of it, including the 2019 price list nobody deleted and the folder where someone saved customer CNIC scans. Most small companies have never tidied their file permissions, because until now nobody was reading everything.

Third, the agent is only as useful as the systems it can reach. It can work in Gmail and Sheets on day one. It can't check stock in your POS, raise an invoice in your ERP or look up an order in your custom system unless those systems have a connection it can use. For most businesses we work with, the real work happens in exactly those systems.

What should you do now?

You can't buy coworker agents today unless you're in the preview, but the preparation is the same whichever vendor's agent you end up using.

Clean up access. Go through your shared drives and remove "anyone in the company can view" from folders that shouldn't have it.
Write down three repetitive jobs that happen every week and have a clear right answer. Chasing unpaid invoices, preparing a weekly sales summary and booking supplier meetings are good candidates.
For each job, list the systems involved. If one of them is a spreadsheet on someone's laptop, that's the first thing to fix.
Ask your software providers whether they have an API or an MCP connector. If the answer is no, an agent can't work with that system.
Decide who approves what. Anything that sends money, sends a message to a customer or changes a price should need a person's sign-off at first.
Check your Workspace plan, and watch for the agent in your admin console. Start with one low-risk job when it arrives.

Should you wait for Google or build your own agent?

It depends where the work lives.

If the job is mostly email, documents, spreadsheets and calendars, wait. A general agent bundled with software you already pay for will be hard to beat on price.

If the job depends on your own software (order status from your POS, FBR or ZATCA invoices from your ERP, stock across branches) then you need the connection built either way. You can build a connector so a general agent like Gemini can use your system, or build a focused agent that does one job inside it. A connector is usually the smaller project, and it keeps working if you change AI vendors later.

What we wouldn't do is sign a large AI contract this month on the strength of a launch event. The product has no public price and no release date.

What else happened in AI agents that week?

The same week, Sierra and Meta announced the Personal Agent Protocol with partners including Shopify, Stripe and Walmart, and published a first draft on 9 October. It's a proposed open standard for how a person's AI agent signs in to a business and acts for them, so the business can tell that the agent represents a real customer.

Put the two together and you can see the direction. Agents are getting identities inside companies, and separately, a way to identify themselves to other companies. If you run an online store or a customer portal, expect some of your "visitors" to be agents acting for customers. We wrote about the shopping side of this in AI shopping on Shopify.

Key takeaways

Google Cloud announced the Gemini agent on 8 October 2026. It takes a goal, works across Workspace, Microsoft 365 and Slack, and can run for hours or days.
Coworker agents get their own email, calendar, storage and directory entry, and their actions are logged under their own name.
It uses Gemini and Claude models and connects to other software through built-in connectors and MCP.
It's in limited preview, with no public release date. Google says there's no additional charge for Gemini Enterprise customers, but hasn't confirmed what's included.
Admins can set hard spending caps per project.
For most businesses, the preparation is the same: tidy file access, pick a few clear jobs, and make sure your own systems can be connected.

Frequently asked questions

What is Google's Gemini agent?

It's an AI agent for work, announced by Google Cloud on 8 October 2026, that takes an objective and completes the steps itself across Google Workspace, Microsoft 365, Slack and connected business software.

Does the Gemini agent really get its own email address?

Coworker agents do. Each one is set up with its own Workspace account, including a Gmail address, a calendar, Drive storage and a company directory listing.

Is the Gemini agent free?

A Google spokesperson said it's available at no additional charge where Gemini Enterprise is available, including through Workspace business and enterprise subscriptions. Google hasn't published a price list or said whether every feature is included.

When will the Gemini agent be available?

A limited number of companies have it now. Google says wider availability is coming soon for select Workspace Business and Enterprise plans, without giving a date.

Can the Gemini agent use Claude?

Yes. Reports say it routes work between Google's Gemini models and Anthropic's Claude models, and that you can choose the model or let it choose.

Will AI coworkers replace employees?

The announcement doesn't answer that, and we'd be guessing if we did. What the product is built for is taking over defined, repetitive tasks under limits a person sets, with a person approving anything risky.

Can the Gemini agent work with my ERP or POS?

Only if there's a connection. It has built-in connectors for well-known business software and supports MCP for everything else, so a custom system needs an API or an MCP connector first.

Want your systems ready for AI agents?

We build POS, inventory and ERP software for businesses in Pakistan, the UAE and the wider GCC, and we build the APIs and connectors that let AI agents work with those systems safely. If you'd like to know what it would take to connect yours, send us your setup on WhatsApp and we'll come back with a written scope and a fixed quote.

This post is based on news coverage of a product in preview. Features, pricing and availability may change before general release, so check Google's own announcements before you plan around it.