Yesterday, 8 October 2026, Google told businesses it has a Gemini agent that doesn't just answer. You give it an objective. It plans the work, uses tools, and can sit inside the software you already run. TechCrunch's write-up is the cleanest account I've read. Sundar Pichai said nearly 90 percent of the Fortune 100 use Gemini Enterprise. Thomas Kurian described an agent you hand a goal to, not a single instruction.
That is a real shift. It is also a description of life inside companies with security teams, spend controls, and someone whose job title includes the word governance. If you employ twelve people in Goodwood, don't hear "90 percent of the Fortune 100" as advice.
What the announcement actually includes
From the reporting, a few details matter more than the keynote adjectives.
- It starts with businesses, not the public chat app
- It can connect to Google Workspace, Microsoft 365, Slack, and a list of heavier data tools most small firms will never open
- Google says it can have its own Workspace account, its own email address, and an audit trail in the agent's name
- You can watch a task inbox: what it planned, what it delegated, what it did
- There are spend controls, because an agent that tries things will also spend money trying things
Read that list as a warning label, not a shopping list. An email address in your company's name, used by a model, is a staff member you didn't interview.
What I'd do this month instead
Finish one job you can explain to a new employee in five minutes. The enquiry reply. The quote chase. The document nudge. The steps are in how to integrate AI into the first job. An agent is a poor first project because its whole trick is deciding the next step. You want that trick after you've seen the next step done correctly a hundred times, not before.
| Situation | A narrow automation | A general agent |
|---|---|---|
| You can name the trigger and the message | Yes | More machine than the job needs |
| The next step depends on judgment | Hand it to a person | Only with a log you'll actually read |
| The firm has never written the process down | Write it down first | It will invent a process. You may not like it |
| A client will see the output | A person approves, until the job earns it | Same rule, higher stakes, because it can touch more tools |
An agent with access to the inbox and no written rule about refunds will eventually issue one. Not out of malice. Out of helpfulness.
When it would be reasonable to look
You've got two or three workflows running. A person checks exceptions. The accounts are in the company's name. You can say, out loud, what the agent is not allowed to tell a client. At that point a tool that can hold an objective is interesting, because the objective is specific: "chase these five stale proposals using the template, and stop if anyone replies with a question."
That's also the shape of an AI operating system. Memory and rules first. A keen worker second. If you want to know whether your firm is at "one job" or "ready for a worker," the discovery call is the honest half hour. Bring the job. Leave the Fortune 100 at the door.
Questions people ask
What did Google actually announce?
On 8 October 2026, at a Google Cloud event, Google launched a Gemini agent aimed at businesses first. It can be given an objective, plan the work, and connect to tools such as Google Workspace, Microsoft 365, Slack, and others. TechCrunch reported it, and the Fortune 100 adoption figure in the room was Sundar Pichai's, not an independent count of small firms.
Should a small South African firm turn it on next week?
Only if you already have one workflow running, written rules for what it may say to a client, and someone whose job it is to read the log. Otherwise you're adding a fast intern to a firm that hasn't agreed the process.
Is an agent the same as an AI operating system?
An agent is a worker that can take a goal and use tools. An operating system is the memory, the data, and the rules around whoever does the work, human or not. You can add an agent later. You can't skip the rules.
What's the risk?
It will do the wrong next step with confidence, in your company's name, inside tools your staff also live in. Spend caps and an audit trail help. They don't replace a narrow first job.