AI Agents at Work: How Enterprises Are Turning Pilots Into Wins

AI agents have moved from demos to daily work inside some of the world's largest organizations. Here is what the early leaders are achieving and the playbook they share.

Enterprise AI has crossed an important threshold. The question in most boardrooms is no longer whether AI agents can do useful work, but how quickly a company can put them to work across every function.

The evidence now comes from deployments, not demos. Banks, energy companies, cruise lines and creative agencies are reporting measurable gains from agents that research, draft, analyze and act. Their experience adds up to a practical playbook for any large organization.

From assistant to agent

The first wave of generative AI answered questions. The current wave completes tasks. An agent can read a stack of documents, break a job into steps, call internal systems and return a finished deliverable for a person to review.

Both Google and Anthropic have built their enterprise offerings around this shift. In October 2025, Google Cloud introduced Gemini Enterprise, which Google Cloud CEO Thomas Kurian described as a single platform that brings together Gemini models, a no-code workbench for building agents, prebuilt agents for tasks such as deep research and connections to company data in Google Workspace, Microsoft 365, Salesforce and SAP.

Anthropic offers Claude through its Enterprise plan and developer platform. In June 2026 it introduced Claude Tag, an AI teammate that works inside Slack: employees assign a task, and Claude breaks it into steps, completes them and delivers the result to the team.

Both platforms also give IT teams a central view of the agents running across the business. Gemini Enterprise, for example, includes governance tools to visualize, secure and audit every agent, and Google says it is supported by a partner ecosystem of more than 100,000 firms. That shared foundation is what allows a company to grow from a handful of agents to hundreds.

Results from the field

The most persuasive case for agents is what customers themselves report. A sample of published results:

• Norges Bank Investment Management (NBIM). The manager of Norway's $1.7 trillion sovereign wealth fund rolled out Claude after a two-week pilot and reached more than 600 active users across all departments within two months. Internal surveys show employees saving about 20% of their time each week on analytical and operational tasks.
• AES. The global energy company built a three-layer, multi-agent system on Claude through Google Cloud to review safety documentation. AES reports that its audit reports are now generated 96% faster, with turnaround falling from about two weeks to roughly one hour.
• TRY. Norway's largest communications group reports a 30% reduction in time spent on routine tasks and 40% faster proposal development, with more than 50 use cases across its creative, strategy and technical teams.
• Macquarie Bank. At Google Cloud Next in April 2026, Google reported that Macquarie had reclaimed more than 100,000 hours of team time with Gemini Enterprise.
• Virgin Voyages. The cruise line now runs more than 1,000 agents and has cut campaign creation times by 40%, according to Google.

What these examples share is scope. None of them is a single chatbot. Each is a portfolio of agents attached to real workflows, with people reviewing and approving the output.

Agents across the business

The gains are not limited to one department. Google's customer announcements show agents at work in sales, service, marketing and operations:

• Customer service. Commerzbank's Gemini-powered assistant, Bene, has handled more than 2 million chats and resolves 70% of inquiries, according to Google.
• Commerce. Klarna used Gemini and Veo to create dynamic, personalized lookbooks and reports a 50% increase in orders.
• Marketing. Swarovski reports email open rates up 17% and campaign localization running 10 times faster.
• Insurance. Signal Iduna now delivers answers to insurance questions 37% faster.
• Advertising. WPP produces an AI-led campaign every four days at twice the previous speed.

The breadth matters. When agents prove themselves in one function, the same platform, data connections and governance can carry them into the next, which is how early pilots become enterprise programs.

Adoption is spreading quickly

The pace of uptake is notable. At Next '26, Google said that nearly 75% of Google Cloud customers use its AI products and that paid monthly active users of Gemini Enterprise grew 40% quarter over quarter in the first quarter of 2026.

Individual companies are moving just as fast. Google reported that KPMG reached 90% employee adoption and built more than 100 agents in its first month, while GE Appliances has deployed more than 800 agents.

Those figures point to a broader pattern. Once employees can build or request agents for their own work, the number of use cases grows on its own. As TRY's chief commercial officer, Kaare Øystein Trædal, put it: "So many agents have been created across the company" that the team is now organizing them with a shared naming convention.

What the leading deployments have in common

Across these companies, four practices stand out.

They start with high-volume knowledge work

Research synthesis, document review, proposal writing and campaign production are natural first targets. The work is frequent, text-heavy and easy to measure, so gains show up within weeks rather than quarters.

They connect agents to real data

NBIM connected Claude to Snowflake and other internal systems, which let teams build dashboards with AI insights included. An agent with access to a company's own information produces work that is specific rather than generic.

They invest in people as much as software

NBIM trained 50 employees as AI Ambassadors to help colleagues adopt the tools. That kind of internal network turns early enthusiasm into lasting habits, and it gives every department a local expert to call.

They keep humans in the loop

In every case above, people review, refine and approve the output. The agent supplies volume and speed. The employee supplies judgment, context and the relationship with the customer.

How to scale from pilot to portfolio

For leaders planning their next phase, the published deployments suggest a clear sequence:

1. Pick three workflows with clear metrics. Hours per report, days per proposal or turnaround per review make progress easy to see and easy to share.
2. Run a short pilot. NBIM's two-week pilot is a useful model: long enough to learn, short enough to keep momentum.
3. Connect the data early. The value of an agent rises with the quality of the context it can reach.
4. Name champions in each function. Trained ambassadors spread good practice faster than any central memo.
5. Publish wins internally. Sharing hours saved and quality gains invites the next team to try.

The sequence also creates a feedback loop. Each successful workflow produces a reusable pattern, from prompts and data connections to review steps, that the next team can adopt in days instead of weeks.

Scaling agents in your organization

AI agents are already producing measurable results inside large organizations, from a sovereign wealth fund to a cruise line. The pattern is consistent: focused use cases, connected data, trained champions and human review.

The practical opportunity is to treat agents as a growing portfolio rather than a single project. Start where the work is frequent and measurable, celebrate early wins and let demand from teams guide the next wave. The companies above show that the payoff can arrive in weeks, not years.

More insights from Enterprise Innovation Center