Google's Gemini Agent Gets a Workplace Identity and Email

Google just gave a piece of software its own corporate email address. On October 8, at its "Gemini at Work" event, Google Cloud rolled out what it calls the Gemini agent — and unlike the chatbot version of Gemini you might use on your phone, this one isn't waiting for you to type a prompt. You give it a goal, and it figures out the steps, picks the tools it needs, and gets on with the work.

From answering questions to doing the job

The pitch is simple: instead of opening five different apps to pull a report together, you tell the Gemini agent what you want done, and it plans the work itself. It can break a big task into smaller pieces and hand them off to other, more specialised AI "subagents" — what the industry is now calling agentic AI, meaning software that acts with some independence instead of just responding to a single command. The agent runs inside the Gemini Enterprise app and Google Workspace, and notably also on rival platforms like Microsoft 365 and Slack. It can reach into a company's internal systems too, the kind of databases and tools that used to require a human login to touch.

Google is also opening the door on which AI model actually does the thinking. The agent can route a task to whichever model suits it best, and by default that now includes Anthropic's Claude alongside Google's own Gemini models, with more third-party models promised later. CEO Sundar Pichai said close to 90% of Fortune 100 companies already use Gemini Enterprise in some form — that installed base is what Google is now trying to turn into paying agent users.

An identity, an inbox, and a lot riding on the guardrails

The detail that stands out is the identity. Each Gemini agent gets its own distinct login, its own permissions, and its own email address, so it can send and receive messages, request approvals, and leave a paper trail like any employee would. Google says this comes wrapped in an audit trail, sandboxing (running the agent in an isolated, locked-down environment so a mistake can't spread to the rest of the system), spend caps, and admin-set policies on what the agent is and isn't allowed to touch.

That caution looks deliberate, because it's catching up to a real problem. Google's own red-team testing earlier this year showed Gemini-style AI breaking into three real companies' systems during a security exercise nobody expected it to win. Giving an autonomous agent standing access to internal tools only raises the stakes if those guardrails slip. Industry-specific versions are reportedly already lined up for financial services, legal, healthcare, government and retail, each with its own tuned skills, so this isn't staying a general-purpose experiment for long.

What it means if you run a business in India

Google is launching this for businesses first, and Indian companies are squarely in that pool. India has been one of Gemini Enterprise's fastest-growing markets, and plenty of mid-size IT, BPO and fintech firms have already been experimenting with AI copilots for over a year. An agent with its own inbox and system access is a different proposition though: it means a piece of software processing customer data, drafting documents, or touching HR and finance systems with real autonomy. That runs straight into India's Digital Personal Data Protection (DPDP) Act, which holds companies accountable for how personal data gets processed, by whom, and with what consent — rules that don't bend just because the "who" is an AI agent instead of an employee. We covered what the DPDP Act actually requires in more detail, and it's worth a read before any Indian business hands an agent admin-level access.

There's an upside for Indian IT services firms too: if clients start expecting agentic AI baked into their workflows, that's a new line of work to build and sell, not just a threat to existing automation contracts.

• Available now in private beta; general availability expected around end-October to early November 2026
• Works inside Gemini Enterprise, Google Workspace, Microsoft 365, Slack, and the command line
• Routes tasks across models, including Google's own and Anthropic's Claude
• Comes with per-agent identity, audit trails, sandboxing, and spend caps
• Industry-specific agents for finance, legal, healthcare, government and retail are planned next

An agent that plans its own work, picks its own tools, and sends email under its own name isn't just automation with a new coat of paint — it's a new kind of coworker, and workplaces haven't finished deciding how much to trust one yet.

The real test won't be the demo stage. It'll be the first time one of these agents makes a judgment call nobody signed off on — books a vendor, emails a client, or touches a system it technically had permission for but shouldn't have used. Google is betting its governance layer holds up before that happens. Every company that turns this on is making the same bet, with its own name attached.