• by Laurie Sullivan , 3 hours ago

Google Cloud outlined its vision for work that starts in a prompt window, where a Gemini agent routes each task to the appropriate model to get it done while keeping costs under control.

Announced at Google Cloud's "Gemini at Work" 2026 event today, Honeywell Technologies Chairman and CEO Vimal Kapur explained in a video on Thursday how the company has integrated Gemini Enterprise agentic workflows directly into its Honeywell Forge IoT platform.

It uses Gemini to learn about its client's system to help predict failures and identify unplanned outages to turn them into planned outages.

Other early customers of Gemini agent were sportswear brand On, Shopify and PayPal. Whereas Shopify blends frontier models for millions of merchants to unlock data and drive sales, PayPal routes 10 million multi-model requests weekly. On, a premium sportswear brand, tested this new capability to accelerate its speed-to-market. 

Google Cloud CEO Thomas Kurian in a blog post published Thursday wrote that users will give the Gemini agent "objectives, not instructions" via an API.

"You delegate an outcome and come back to finished work," Kurian wrote. "For an agent to do that, it has to be connected to your personal workflows, your systems of record and your enterprise controls."

He added that the agentic capabilities in Gemini are built around core architectural principles such as "unified agent," omnipresent access, "persistent execution," "multi-agent orchestration," and "deeply contextual," and "model choice flexibility."

Gemini will pull information company-wide to complete tasks. The agent understands the business model and protocols, everything from pricing to product portfolios to departmental processes. Tools that connect it to the company's systems, skills that teach it how work gets done, and context that allows it to remember are three primary principals that make a company unique and agentic automation possible. 

According to Google, nearly 80% of all Google Cloud customers today use its AI products, and nearly 90% of the Fortune 100 use Gemini Enterprise. 

The RealReal, an online marketplace for authenticated luxury resale items, for example, supports luxury consignment by deploying "Ask TRR," a conversational shopping concierge built with Gemini Enterprise.

Resale shoppers have navigated and searched a massive, shifting catalog using natural language instead of rigid filters, which will boost user engagement across all 45 million members while making it effortless to discover and secure rare luxury pieces before they sell out.

Kurian explained how marketers can access Google's agent from any device and channel including Google Workspace, Microsoft 365 and Slack as well as third-party applications. The Gemini agent can create a roster of sub-agents to carry out work and multi-step tasks.

Historically, using a Google AI tool meant the user could only use Google models. Now, the Gemini agent can route tasks to non-Google models, starting with Anthropic's Claude. Marketers are not locked into one tech ecosystem and can use any AI tool is best for their specific project.