Artificial Intelligence

What Gemini Enterprise Can Run for You Today

Introduction

Most of the businesses we talk to that pay for Gemini use it the way they use any chat assistant. Someone opens a window, asks a question, copies the answer somewhere else, and closes the window. The work starts and ends with a person typing.

On October 8, at its Gemini at Work event, Google announced the Gemini agent opens a new window , which it calls “your new single, universal agent for work.” You give it an objective, it works out the steps, and it can keep going for hours or days inside Gmail, Drive, Docs, Sheets, Slides, Chat and Calendar. As of late October, it is in private preview, and Google has not given a date for the wider rollout.

What gets less attention is that a smaller version of this idea already ships in Gemini Enterprise: agents you build once, connect to your company’s data, and schedule to run without anyone opening a chat. A business that sets a few of those up now can get some of the recurring work off its plate this quarter, and find out where its data and permissions aren’t ready while the stakes are still small. In this post, we’ll cover what the October announcement promises, what Gemini Enterprise can already run on a schedule, four recurring jobs worth handing over, and where a person still has to step in.

We covered the same copy-and-paste loop, and Anthropic’s answer to it, in Claude in Google Workspace: What to Know opens a new window . This post is about what Google ships for the same problem.

What the October announcement promises

Google’s pitch for the Gemini agent is that you “give it objectives, not instructions.” According to 9to5Google opens a new window and VentureBeat opens a new window , you will be able to call it with @Gemini from Gmail, Docs, Sheets, Slides and Chat. It splits a job across sub-agents (smaller agents it creates for each part of the task and runs in parallel), picks between Gemini and Anthropic’s Claude models depending on the task, and remembers how your team does things from one job to the next. Google’s own example is asking Gemini to set up a meeting with the usual team of regional event leads, without naming anyone: Gemini works out who they are from the membership of a chat space and the thread from the last event, checks their calendars, and starts the email thread to find a time itself.

The part with the most potential for a small team is what Google calls coworker agents. You describe a role, such as the events coordinator in Google’s example, and Gemini creates an agent with its own company email address, calendar and Drive storage. Colleagues assign it work by adding it to a Chat space, mentioning it, or tagging it in a document comment, the same way they would a person. Google says agents can only open files that were explicitly shared with them, and that every action an agent takes is logged against its own identity.

Google told VentureBeat the agent will be included at no additional charge wherever Gemini Enterprise is available, including through existing Workspace business and enterprise subscriptions. VentureBeat notes the open questions: whether coworker accounts need their own licenses, how long-running tasks and sub-agents are billed, and how many coworkers you can have. Google says pricing details will come closer to release, so treat “no additional charge” as a statement about access rather than about your bill.

What already runs today

While the Gemini agent stays in preview, the pieces you can use now live in Agent Designer opens a new window , the no-code agent builder in Gemini Enterprise. You describe what the agent should do, choose which of your company’s data sources it can read, and Gemini writes the instructions. The sources are Gmail, Drive and whichever connectors your administrator has turned on. A connector is the link between Gemini Enterprise and an outside tool such as Jira or your CRM. Google’s own example prompt on that page builds an agent that drafts follow-up emails for new leads from a Salesforce connector. For bigger jobs, you can open a visual builder and split the work across sub-agents, each with its own instructions and data.

One thing to check before you plan around any of this: these agents are a Gemini Enterprise feature, and Gemini Enterprise is its own subscription with Business, Standard, Plus, Pay-as-you-go and Frontline editions. The Agent Designer help page linked above is for the Business edition, and Google’s cloud documentation covers the other four. If your company only has the Gemini that comes with a Workspace plan, ask your administrator which edition, if any, you are on.

The feature that turns this from a chatbot into something that works while you’re away is scheduling. An agent can run hourly, daily, weekly, monthly or yearly, in your time zone, with a saved prompt. Each run shows up in your chat history opens a new window like any other conversation. What happens when a scheduled run needs to reach someone else depends on your edition. On the Business edition, the help page opens a new window says work that only affects you runs on its own, and “any action involving other people will pause for your review and approval.” The agent drafts the email, and you decide whether it goes out. The documentation for the other four editions is stricter: “scheduling agents is not supported for actions involving other people.” On those editions, a job that ends in an email to a client belongs in a workflow, which we cover next, with a person approving the send.

For the Standard, Plus, Pay-as-you-go and Frontline editions, Google’s documentation now calls the same tool Workflow Builder opens a new window (the rename from Agent Designer is in the September 3 release note opens a new window ). On those editions it also includes workflows, which chain AI steps with fixed rules and human sign-off. You can add review, approval and information-request steps wherever you want a person in the loop, and Google requires human review whenever a workflow adds or updates data through a connector. Workflows can be started by hand, on a schedule or by an event, and Google’s documentation uses three examples that are close to real small-business work: an onboarding checklist that runs across Gmail and Calendar, a report that the AI drafts and a person signs off before it is published, and incoming leads routed by region and budget. Google’s editions page does not say whether the Business edition includes workflows, but Business users have the paused scheduled run for the same purpose, so the four jobs below work on either path.

Four recurring jobs worth handing over

We wrote about the general bar an agent has to clear in When Does Your Business Need an AI Agent? opens a new window . The jobs that suit a scheduled agent in particular share three traits, which overlap with the characteristics we described in Automating Repetitive Work with Skills and MCP opens a new window . They happen on a predictable rhythm, the information already lives in systems the agent can read, and a person can check the result in a few minutes. Here are four we would set up first.

A Monday pipeline brief

If your customer relationship management (CRM) system is connected, a weekly agent can read the deals that moved, stalled or slipped a close date, cross-check them against recent email threads with those customers, and leave a one-page brief in your chats before your Monday meeting. Because the brief only goes to you, it runs without an approval step on any edition. The value is less in the summary and more in the cross-check, because the deal that looks healthy in the CRM and has gone quiet in email is the one nobody mentions until it’s lost. Keep the brief to a fixed structure that you write once, so the agent produces the same shape every week and you can tell at a glance what changed.

A client status report you only have to approve

Many service businesses send a weekly or monthly status update to each client, and someone spends a Friday afternoon stitching it together from tickets, shared docs and email. An agent with access to the client’s Drive folder and the project tracker can draft that update on a schedule. Because sending it involves another person, it needs the approval step from the previous section: a paused scheduled run on the Business edition, or a workflow with an approval step on the others. Your Friday job shrinks from writing the report to reading it, which is the same shift we saw when we automated our own quarterly maintenance reports opens a new window . Give the agent a past report you were happy with as a reference file, and ask it to flag anything it couldn’t verify instead of filling the gap.

A follow-up sweep

A daily agent that reads your inbox can list the threads where you promised something and haven’t replied, the customers waiting on an answer for more than two days, and the invoices someone asked about. It can draft the follow-ups too, and each draft waits for your approval, as a paused run or a workflow approval step, before anything is sent. A missed follow-up usually costs more than the few minutes it takes to approve a draft, and catching them by hand is exactly the kind of tedious scan people skip on a busy week.

A shared “how we do things” agent

Most small businesses have a Drive folder of policies, price lists and process documents that few people open. An agent built over that folder and shared with the team answers questions like “what’s our refund policy for annual plans” or “who approves a discount over 15%” with a link to the document it used. This one doesn’t need a schedule, but it is the agent that pays off the effort of organizing Drive, which you will need for everything else on this list.

Where it needs a human

The limits are documented, and a few of them will bite if you don’t plan for them.

Schedules run on your credentials, and Google’s documentation says they must be refreshed every 14 days or they fail. A weekly report that quietly stops in week three is worse than no report at all, so put a recurring reminder on someone’s calendar to refresh schedules, and check that each scheduled agent produced its last run. Runs can also start up to five minutes late, and the Business edition help page caps each person at 20 schedules.

Agents can only use the data sources your administrator has enabled. If the CRM isn’t connected, the pipeline brief can’t be built, and no prompt will fix that. Decide which connectors the business actually needs before anyone starts building.

Sharing an agent shares its access. Google’s help page warns that people you share an agent with can query its data sources and files, even if they don’t have direct access to those files. An agent built over the finance folder and shared with the whole company has effectively shared the finance folder. Build shared agents over content you’d be comfortable sending to everyone who can use them.

The approval step is only as good as the person approving. When a draft arrives every week and is usually fine, people stop reading it carefully. Spot-check one run in five properly, against the source data, and you’ll notice drift before a client does.

Conclusion

When the Gemini agent reaches Workspace plans, it will start from whatever your company has already organized. It can only open files explicitly shared with it, it learns your processes from what it can read, and it is governed by the permissions and connectors you have set. A business that has used the current agents for a few months will have answered the questions the new one depends on: which data sources are connected, which folders are clean enough to share, who approves what, and which recurring jobs are worth delegating at all (we wrote about how to pick those in Finding the Right Problems to Solve with AI opens a new window ).

The scheduled agents are worth building now for their own sake, and they are also one of the cheaper ways to find out where your data, permissions and approval habits aren’t ready yet. Pick one recurring job from the list above, build the agent this week, and run it alongside the person who does the job today for a month before you let it run on its own.

We help teams put AI to work on the tools they already use. Not sure which recurring job to hand over first, or how to connect the data it needs? Let’s talk opens a new window .

Our AI Services

Turn your data into a competitive advantage

View AI Services opens a new window