From Levi's to EY: How Major Firms Are Deploying AI Agents to Redesign Work

The Next Enterprise AI Frontier: Agents That Run Processes, Not Just Assist People

Artificial intelligence inside major corporations is pivoting from a personal productivity tool—drafting emails or summarizing meetings—to something far more structural. A fresh wave of AI agents, highlighted by Microsoft in its latest product showcase, now executes multi-step business processes end-to-end, often without continuous human direction.

Microsoft's Copilot Cowork platform lets users task an AI with an entire workflow: analyze a dataset, then automatically draft an explanatory email to stakeholders. A different agent, Microsoft Scout, works in the background monitoring projects and acting within pre-set permissions, edging toward the concept of a permanent autonomous digital colleague. Behind the scenes, over 30 million paid Microsoft 365 Copilot licenses have been deployed, and the number of customers with more than 50,000 seats has surged more than sevenfold year-over-year, the company said.

Concrete examples from early adopters punctuate the shift. Power-management multinational Eaton has rolled out a quality agent that digests incident reports, spots patterns, and suggests root causes across supplier and production data, already analyzing some 5,000 reports. At Levi Strauss & Co., a finance-focused agent sifted through 1,100 operational procedures in a single day, cataloguing roughly 18,000 individual tasks—spade-work that finance VP Lisa Sterling says was practically impossible manually. EY’s Autonomous Procurement Agent has supported more than 200 transactions since an October 2025 pilot, handling low-risk purchasing from requisition to purchase order while keeping critical decisions human-led. S&P Global Commodity Insights embedded an agent in Microsoft Teams that slashes data extraction time by 95% and comparative analysis by 98%, giving traders and investors faster access to complex energy and commodities intelligence.

Meanwhile, Premera Blue Cross employees built more than 900 specialized agents themselves via Copilot Studio; one contract-attachment agent cut a 30-to-45-minute manual task to roughly three minutes. HR bots at Kantar now complete employment verification letters in about two minutes instead of three days, resolving 40% of HR requests with a goal of 95% by year-end. Even Microsoft's own product engineering teams used AI agents to accelerate development of Copilot Cowork with a tiny team of nine, and its cloud supply chain unit reduced cycle times by 75% after redesigning processes before automating them.

Inside the Agent Revolution at Levi’s, Premera Blue Cross, and S&P Global

Why This Is Not Just the Next Iteration of Office Copilots

The examples shared by Microsoft mark a deliberate pivot: the conversation is no longer about saving a few minutes per task for a knowledge worker but about reconstituting entire workflows. Where earlier AI assistants largely generated text or answered questions, agents now chain together data extraction, analysis, drafting, and even procurement decisions. This matters because the productivity uplift shifts from incremental to structural—when a contract that took 45 minutes now takes three, capacity is freed for higher-value work rather than merely speeding up the old way. However, the source is Microsoft’s own communications, meaning the examples are curated to showcase successful deployments; we should expect variation in outcomes at other organizations.

Democratizing Agent Creation—and the Governance Challenge It Brings

Premera Blue Cross’s approach, letting employees create their own agents, suggests that adoption can scale faster when the people doing the job design the tool. But more than 900 specialized agents also raises questions about consistency, security, and overlapping functionality. The emerging frontier is not just building agents but governing a fleet of them. Microsoft’s own “Frontier Tuning”—imbuing HR agents with specialist knowledge—claims five times the success rate of generalist models, but the technique still relies on the quality of the embedded rules and context. For organizations with less mature data practices, a scattergun agent strategy may create more chaos than efficiency.

What the Microsoft Ecosystem Concentrates, and What It Locks In

Nearly every example runs on Microsoft 365 Copilot, Copilot Studio, or Teams. This concentration carries both an advantage—deep integration with widely used productivity tools—and a risk of vendor dependency. Companies building critical workflows on a single platform’s agent architecture may find switching costs prohibitive down the line. The article underlines that Microsoft itself redesigned its cloud supply chain processes before deploying agents; that principle of “simplify first, then automate” is essential but also harder to execute when the automation blueprint is tightly coupled to one vendor’s stack.

The Data Point That Signals a Tipping Point

Microsoft reports that nearly half the work done in multi-step Copilot Cowork flows involves analytical components, not just content generation. If accurate, it suggests that AI agents are moving from periphery tasks into the analytical core of business operations—where they can influence significant decisions. This shift, if sustained, could redefine which roles are reserved for humans and which become agent-managed. For now, the retention of human oversight on critical transactions at EY and the deliberate “augment, don’t replace” framing at Eaton indicate that the most thoughtful adopters are pursuing hybrid models.

Where Business Leaders Go From Here

For business and technology leaders watching this evolution, the early deployments offer clear cross-industry lessons:

• Map and simplify first, then automate. Microsoft’s cloud supply chain team cut cycle times 75% only after redesigning six end-to-end workflows. Automating a broken process just accelerates inefficiency; start by auditing existing workflows as Levi’s did with its 18,000 financial tasks.
• Let domain experts build agents—within guardrails. Premera Blue Cross gained speed by empowering employees to create 900 agents. Consider a center-of-excellence that provides governance while frontline teams design solutions for their own pain points, replicating the 95% faster HR letters at Kantar.
• Retain human judgment on high-stakes decisions. EY’s procurement agent handles routine buys but escalates critical choices. Define clear categories where automation stops, as Eaton does for quality root-cause analysis—the agent suggests, the human decides.
• Watch for vendor lock-in as agent fleets expand. If your organization standardizes on one platform’s agent framework, the cost and complexity of future migration rises. Build interoperability requirements into agent design now, rather than retrofitting after hundreds of agents are live.
• Measure the shift in time allocation, not just task speed. Microsoft’s sales team doubled customer-facing time from 25% to 50% because agents absorbed admin work. Track where freed capacity actually goes—if it doesn’t move toward revenue-generating or strategic activity, the structural promise isn’t being realized.