In the past two years, terms like "AI agent" and "intelligent agent" have been overused. But in recent weeks, some very concrete things have been happening—not PowerPoint presentations, but actual deployments. Looking at these things together reveals a clear thread of change.
The First Shot in the Audit Industry: EY Integrates 130,000 Auditors into an AI Assembly Line
On April 7, EY announced in London that its global audit platform, EY Canvas, has completed the full integration of enterprise-level agents. This is not a pilot — it is a full-scale deployment targeting 130,000 auditors, covering 160,000 audit projects, and spanning over 150 countries.
The EY Canvas platform processes over 1.4 trillion rows of journal entry data annually. The multi-agent framework integrated this time is built on Microsoft Azure, Foundry, and Fabric, and is expected to support all end-to-end audit activities by 2028.
130,000 audit professionals, 1.4 trillion years of data rows processed, 150+ countries and regions covered
What deserves attention is not the scale, but the deployment method. EY has embedded multiple AI agents directly into existing audit workflows, rather than equipping employees with a separate "AI assistant." The agents are responsible for orchestrating complex tasks, retrieving the latest audit standards in real time, and dynamically assessing risks; humans are responsible for judgment, questioning, and final confirmation.
EY positions itself as "Customer Zero" — first applying this system to itself, and after verifying it, rolling it out to audited companies globally. Many large companies are now following this logic.
Another Path for IT Service Providers: DXC and ServiceNow's "Use It Yourself Before Selling It" Model
Also on April 7, IT outsourcing giant DXC Technology announced a multi-year partnership agreement with ServiceNow, under which DXC will become the first "Customer Zero" customer of ServiceNow's newly released Core Business Suite globally.
In simple terms: DXC first applies ServiceNow's AI automation tools to its own internal global business service system, and after making it work, packages this proven solution and sells it to other enterprise customers. DXC has over 1,800 ServiceNow consultants, who will no longer sell "theoretical solutions" but rather things they have used themselves.
"Global enterprises face pressure to shift from AI experimentation to true execution—which is difficult when your operations are complex and distributed. DXC chose to act first." — Josh Kahn, Senior Vice President of Core Business Workflows at ServiceNow
This model has some reference value. If consulting firms and integrators have not used this tool themselves, their persuasiveness in selling it to clients is very limited.
Domestic: L3-level intelligent agents officially commercialized, manufacturing sector leads the way
On the domestic front, the Tencent Cloud Developer Community reviewed benchmark implementation cases of the AI agent industry in early April 2026. A noteworthy assessment is that AI agents have moved from the "L2 assisted decision-making" stage to the "L3 autonomous execution" stage, with the number of domestic service providers exceeding 300.
A few specific figures: DeepMiner from Minglue Technology achieves a single-step operation accuracy of 98.9% in industrial scenarios, connecting to over 80 data sources. Among its clients are 135 Fortune Global 500 companies.
Manufacturing has become the most intensive application area. The reason is not hard to understand: tasks on the production line are highly repetitive, data structures are relatively clear, and the margin for error is small. These characteristics both suit AI agent intervention and demand high accuracy. Industrial scenarios are the first to be implemented, followed by finance, retail, and government services.
Why haven't most companies succeeded in this yet?
In its early April report, Gartner presented a somewhat disheartening figure: out of every 50 corporate AI investments, only one delivers transformative value, and only one in five generates measurable returns.
The reason is not that the technology is not good enough. Gartner pointed out several issues:
First, the quality issue of AI-generated content. Employees spend nearly two hours on average handling each instance of "AI-generated garbage output" — correcting errors, fact-checking, and rewriting structure. This time cost is invisible in most ROI calculations.
Second, the misalignment between layoffs and expectations of AI productivity. Many companies thought that adopting AI would allow them to use fewer people, but they ended up finding that people are still needed—just those who know how to use AI—and the recruitment costs are even higher than before.
Third, process expertise is more valuable than AI tool skills. Departments that can redesign workflows using AI are twice as likely to exceed revenue targets. The key is not which tool you can use, but whether you can think through existing processes clearly before involving AI.
Summary
Taken together, these matters indicate one thing: the application of enterprise AI agents has moved past the "demonstration phase," but there is still some distance to go before reaching "large-scale effective application." What stands in between are process redesign, personnel skills, and quality management of AI outputs.
Directions worth continuing to observe next week: whether the other four major audit firms will follow EY's deployment pace, and how domestic ERP platforms such as Odoo and SAP will embed intelligent agent capabilities into existing enterprise management workflows.
Enterprise AI Agent Digital Transformation Intelligent Agent ERP Audit Technology Smart Enterprise Digital Observation Focuses on Enterprise AI Implementation, ERP Modernization, and Digital Management Practices. Daily selects 3-5 valuable industry trends.
