Google brings agentic AI to Gemini, starting with businesses
TechCrunch Sarah Perez ● Covered by 3 sources
Google is turning Gemini into an agent that can do tasks, not just chat. It’s launching with businesses first, and Google says Gemini already has over 1 billion users.
Based on reporting by TechCrunch, Sarah Perez — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Google used a Cloud event on Thursday to show where Gemini is headed next: less chatbot, more worker. The new agent is meant to answer questions, sure, but also to carry out tasks on a user’s behalf from one interface. Google is pitching this as the shift from conversation to action, the point where AI starts taking ownership of the job instead of just talking about it.
The first stop is business, not consumers. That fits Google’s current footprint: Sundar Pichai said Gemini has more than 1 billion monthly active users, and nearly 90% of Fortune 100 companies now use Gemini Enterprise at work. Google wants to use that base to tackle the ugly parts of agent software first — security, scale, performance — before it opens the door wider.
Thomas Kurian, who runs Google Cloud, described the system as one that takes objectives rather than simple instructions. In practice, that means the agent can plan work, use custom skills and tools, and plug into internal systems to finish the task. By default, Google will choose the model it thinks fits best. Users can override that and pick a model themselves, starting with Anthropic’s Claude models, and Google says open source and other private models will be added later.
The agent can work with attachments such as files and folders, and with workstream projects that combine files and skills. It can also connect to a long list of business systems: Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres, Snowflake, and more. Google says it can also operate securely with any Model Context Protocol server inside or outside a company’s network.
There’s a tasks inbox to watch the agent think, hand work to subagents, load special skills, run code, and show progress. Google also says the agent gets its own Workspace account, email address, and context, so it behaves less like a feature and more like a coworker. It can be tagged, emailed, shared with, or added to a group chat, and it leaves an audit trail under its own name. Early testers included On, Shopify, and PayPal, while Google named BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty, and Wesfarmers as Gemini Enterprise customers.
My take — AI-written commentary, not fact-checked reporting
This is the right place for agentic AI to start: inside companies, where the mess is real and the guardrails actually matter. The open-model crowd will hate the default tilt toward Google’s own stack, but at least Google is admitting the hard part is control, not demo magic. Europe should be paying attention too, because the audit trail and model choice story is what regulation keeps asking for, just with more paperwork and less charm.
Read more about this at: TechCrunch