New York City skyline in Manhattan with a prominent central skyscraper, flanked by two tall buildings under cloudy skies

How CAOs are defining the new model for human/agent workforces

Photographic portrait of Myles Corson
Myles Corson

EY Global Financial Accounting Advisory Services, Strategy and Markets Leader

9 minute read 07 Aug 2026

As AI accelerates, finance focuses on enhanced talent, trusted data and reimagined governance. Here’s how Fortune 100 CAOs are thinking.

In brief
  • The apprenticeship model may need to be rebuilt as AI changes how early-career finance professionals develop judgment as well as broader business skills. 
  • In human/agentic workforces, AI governance will need to become as deliberate as financial control itself, with clear owners, risk tiers and accountability.
  • Clear communication stands out as an imperative for managing change and defining a narrative that finance functions can anchor on amid uncertainty.

How EY can help

As AI capabilities rapidly advance, how should the operating model for controllership keep pace? Known for delivering accuracy and consistency but not necessarily for quickly embracing change, accounting and finance functions are challenged to reimagine nearly every facet of what they do and how — and CAOs and controllers at some of the world’s biggest companies must forge roadmaps through uncertainty.

 

Designed for this moment, the EY F100 CAO/Controller Leadership Network (CCLN) brings together peers to apply their collective experience to these complex challenges. Over 20 members of the network gathered recently at the EY wavespace™ in New York for an immersive experience, enabled by technology and designed to spark new thinking.

 

“If folks are spending 80% of time in front of the computer now and 20% as business partners, that will flip,” one controller said. “That’s the gap that must be addressed.”

 

And the implications from that are vast, noted Tiffany McDowell, a People Advisory Services principal with Ernst & Young LLP. “My belief is that we don’t want tech to design our future for us,” she said. “We need to think intentionally about what should be valuable for humans to handle and tech takes the rest.”

 

Measuring consequence against trust

McDowell said that CAOs and controllers should start by defining what work is valuable and how accountability is maintained, and from there, leaders can look at the skills, roles and organizational models to enable those decisions. With that in mind, she introduced a two-by-two framework to help pinpoint what tasks should be handled by agents, humans or a combination of both.


How EY can help

On one axis is how much conviction leaders need in the work, and on the other is the potential consequences from that work. If both are low, agents are likely best suited for the task; if both are high, humans should retain the task. CCLN participants were polled on a number of responsibilities within controllership to map against this rubric.

 

There was strong alignment on activities and decisions that should remain with humans, including approval of material and judgmental journal entries, and on those more suitable for agents, such as recording standard period-end journal entries. (“Don’t ask us about how far we’ve gotten by next year, though,” one participant joked.) 

 

“There’s a split in the top left to bottom right diagonal slant of the framework, and that’s where we need to get governance right as we define the right levels of human oversight and AI assistance,” said Myles Corson, CAO/Controller Program Leader for the EY Center for Executive Leadership (CEL). Activities where there were different views on the balance of human and AI interaction included:

  • Certifying SOX 404 attestation. Risk tolerance differed across member companies, and the CAOs had varying visions of what this certification process would look like under a hybrid human/agentic workforce.
  • Preparing ASU adoption impact analyses. “If it’s about data you don’t have or deeper analysis you need to do, then you’d need a human,” one member said, while another added: “I’m not set up to fully ‘agentify’ it today. But I believe I can get there.”
  • Conducting variance analysis on flagged anomalies. “We have models that do this now,” a CAO said. But another clarified: “The flag of the anomaly could be automated, not the investigation.” Others felt that AI was indispensable for understanding the anomaly and even how it’s communicated by writing the first draft of reports.

Separately, as the cost of the tokens used for AI to process inputs and generate outputs becomes more apparent, another question takes on added importance: Even if a task neatly slots into fully automated agentic processes, is the ROI actually worth it? And how does the operating model need to change to maximize impact? Our discussion centered on three key pillars to consider.

1

Chapter 1

Talent and judgment

CCLN participants feared the erosion of the apprenticeship model in the event that entry-level jobs begin to vanish but the need for the judgment cultivated in the early years of a career must persist. “There’s a value in aggregating the data and more value in understanding it,” said one participant. “My understanding is intensified by building it.” Many members noted that, in their careers, they were promoted based on their abilities in a specialized area of accounting or in a certain tech system, whereas broader and more generalized skills are more relevant.

How do you step into this world of hybrid agentic/human ways of working without creating too much fear? Employees need to take the journey and be empowered to learn and build, even as the required skills rapidly shift. “Accountants have to be better business leaders,” a CAO said. “That’s a tough skill for a lot of them. We have to teach them.” That in turn has a knock-on effect for performance metrics and KPIs.

Actions to consider

2

Chapter 2

Data and trust

One common thread among many controllers: so many important sources of data are inconsistent, with no single source of truth across potentially dozens of ERPs. “We have 50 ERPs, and we’re going through a massive transformation to scale down,” one leader said. “Even when our only source of truth is wrong, we have to treat it as right.” Another participant noted the flip side of this debate: “We just completed our ERP implementation after three years,” showing how addressing legacy tech is a marathon while AI moves at a sprint.

Concerns about auditability also loomed over the discussion. How do you validate that the agents are working effectively and maintain that validation to the satisfaction of auditors? Just managing AI inventories can be a challenge as the rise of “vibe-coding” creates potential shadow AI problems. “It’s an extension of data governance — you almost need an AI dictionary.” And meanwhile, off-the-shelf agents are available everywhere: If you buy, who from? Is it a cloud app? Is it within your existing platforms? At what price?

Actions to consider

3

Chapter 3

Governance and accountability

Overwhelmingly, CCLN members cited the pace of change as a struggle. “Basically our company makes us feel like we’re behind when we haven’t even started,” one said. And governance can get crowded out, amid dozens of other projects. Everyone’s pitching the board while the media narrative makes everyone seem like they’re lagging. “Every one of our executive sessions is: ‘Where are we on this?’” another CAO added.

Amid this scrum, ownership clarity and accountability are prominent question marks. AI efforts must be cross-functional, which blurs the lines on who is doing what. “We don’t live in a static environment, so there are 20 things going on at the same time,” a controller said. Accounting and finance functions do not generate revenue, so they can be overlooked as a priority for transformation, although they can destroy shareholder value if not handled properly. 

Actions to consider

The importance of clear communication

Communication is core to change management, yet controllers may find it a struggle, whether they are talking to the board or to their direct reports and teams. Tim Teagle, EY Modern Client Experience Leader, shared a proven approach to delivering a message crisply and confidently to connect with your audience.

Summary

AI is forcing controllership leaders to reconsider how work is performed, governed and measured in an age of hybrid agentic/human teaming. In an EY wavespace™ session, CAOs and controllers explored how to allocate tasks for maximum impact based on trust, consequences and accountability: by developing talent and judgment, improving data quality and auditability, and clarifying governance for AI initiatives. The path forward requires experimentation, targeted investment and transparent communication with boards and teams about what can be possible and what must be prioritized.

About this article

Photographic portrait of Myles Corson
Myles Corson
EY Global Financial Accounting Advisory Services, Strategy and Markets Leader
Supporter of finance leaders and functions of today and tomorrow. Global citizen. Podcast host. Curious mind. Father of two energetic boys.
Related topics
Assurance
Financial accounting advisory services
Technology

Related articles

How CAOs and controllers are forging the future of finance

From AI in finance to talent upskilling, discover key insights on finance transformation

Myles Corson

EY refers to the global organization, and may refer to one or more, of the member firms of Ernst & Young Global Limited, each of which is a separate legal entity. Ernst & Young Global Limited, a UK company limited by guarantee, does not provide services to clients.