Google is turning Gemini into a persistent workplace agent as Manus and AI-evaluation startup Arena raise major rounds.
Google Cloud’s latest AI move is less about a new chatbot than a digital coworker. Announced October 8, its Gemini enterprise agent can run persistently in the cloud, retain several types of memory and delegate tasks to parallel or sequential subagents. Google says work can continue for hours or days.
The agent connects with Microsoft 365, Slack, Teams, Google Workspace, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres and Snowflake. It can also connect to internal or external servers using the Model Context Protocol, or MCP. Google Cloud CEO Thomas Kurian described the shift as giving agents “objectives, not just instructions.”
Google says each agent gets a dedicated company email address, persistent storage and a Workspace account. Its actions are logged under the agent’s identity rather than a human employee’s. Agent Sandbox and Agent Gateway are intended to provide execution isolation, network controls and policy enforcement.
That design puts permissions and accountability at the center of deployment. An agent that can reach code, tickets, customer systems and databases needs limits as well as an audit trail. Google lists Git and MCP among its connections, making the platform relevant to teams integrating agents with repositories and custom tools. The service is in preview for financial-services and legal customers; government, healthcare and retail are planned next.
Google also says users can select third-party models, initially including Anthropic’s Claude. That could let businesses combine models within a cloud platform rather than commit every workflow to one provider. Google describes spending controls such as multi-model orchestration, smart routing and real-time caps, aimed at managing costs as agents take on longer tasks.
The company says nearly 80% of Google Cloud customers use its AI products, nearly 90% of Fortune 100 companies use Gemini Enterprise, and nearly 500 Cloud customers processed more than one trillion tokens over the past year. These are Google’s figures, not independent measures of agent performance. They illustrate the scale of the enterprise market it is pursuing.
Meanwhile, Manus is operating independently again after its proposed deal with Meta ended. Butterfly Effect, the startup’s parent, said October 8 that it raised more than $500 million in a round led by Boyu Capital and IDG Capital. Tencent, HSG—formerly Sequoia China—and ZhenFund also participated. The company did not disclose a valuation; TechCrunch reported it had recently discussed a $4 billion figure.
Manus also launched Manus 2.0 and Cue, an app that gives agents email addresses, phone numbers, digital wallets and computers. TechCrunch reported the company is continuing to hire domestically and internationally. The funding follows the end of Meta’s proposed $2 billion acquisition and Manus’s return to independent operations.
A separate round underscores investor interest in evaluating AI systems. Arena raised $200 million in Series B financing at a $3.1 billion valuation, according to TechCrunch. Lightspeed Venture Partners and Khosla Ventures led the round. Arena, which began with model rankings, is expanding into AI-safety and alignment evaluation.
Together, the announcements point to investment in more than models: agents that act, platforms that connect them to workplace systems, and services intended to assess their behavior. The contest is increasingly about who controls agents’ access, identities and guardrails—not just which model produces the best answer.

