EY, one of the world’s largest professional services firms, is creating a dedicated unit to get its ballooning artificial intelligence costs under control. The Big Four firm is hiring a head of “agent economics” to lead a new AI value realization office — a function that barely existed as a job title twelve months ago.

The move, reported by Bloomberg on August 10, signals that even the organizations most aggressively deploying AI are now confronting an uncomfortable truth: scaling AI agents is expensive in ways that most business cases didn’t account for.

The Cost Problem Nobody Planned For

Running a simple AI chatbot is cheap. Running an agentic AI system — one that reasons, uses tools, checks its own work, and loops through a task until it’s done — is a different story entirely.

Simple AI interactions cost around $0.04 each. Complex agentic workflows that involve tool use, multi-step reasoning, and iterative loops run closer to $1.20 per interaction. That’s roughly 30 times more expensive, and the costs compound fast at enterprise scale.

EY has embedded AI deeply into its operations. The firm has deployed agentic AI across its global audit practice, covering more than 130,000 professionals. When you’re running millions of AI interactions across that workforce, a 30x cost differential isn’t a rounding error — it’s a budget crisis.

Dan Diasio, EY’s global consulting AI leader, described the intent clearly: the AI value realization office will give leadership “end-to-end visibility of where and how our investments in AI are driving material and tangible impact in our business.”

A New Job Category Arrives

The title “head of agent economics” is new. Six months ago it didn’t exist in any meaningful way. Now it’s a C-suite-adjacent hire at one of the most AI-forward professional services firms in the world.

That trajectory matters. New job titles at firms like EY tend to signal where the broader market is headed. The role of Chief Data Officer didn’t exist twenty years ago either — and now every enterprise of scale has one, or a direct equivalent.

Agent economics is becoming its own discipline because the old frameworks don’t fit. Traditional software has fixed licensing costs. Cloud compute scales predictably with load. AI agents are different: their cost scales with task complexity, model choice, number of tools called, and how often they need to retry. Those variables are hard to forecast and harder to govern without dedicated oversight.

The Pattern Is Industry-Wide

EY is not alone. Across the enterprise landscape, finance and technology teams are waking up to the same problem. AI token costs have emerged as a surprise line item in quarterly reports at companies that deployed agents aggressively in 2025 without modelling the full cost of production-scale operation.

The challenge isn’t whether AI delivers value — for most organizations that have deployed it properly, it does. The challenge is whether the value it delivers exceeds what it costs to run, and whether anyone in the business has clear sight of that calculation.

For many companies right now, the honest answer is no. Agents are deployed, they’re running, they’re doing useful things — but the cost-per-outcome number is unknown, and no one owns it.

What This Means for Business

If EY, with its resources and AI expertise, needed to build a dedicated function to manage AI costs, most businesses without that infrastructure are likely in a more exposed position.

A few things to take from this:

Know what your agents cost to run, per outcome. Not per month, not per seat — per outcome. The unit of measure matters. An agent that costs $1.20 per interaction but reduces a task from 40 minutes to 2 minutes may be extremely economical. One that costs $0.20 per interaction but delivers nothing the business couldn’t already do is expensive at any price.

Model complexity before you scale. Simple prompts and complex agentic pipelines have very different cost profiles. Don’t build your ROI case on prototype cost and then scale the production version.

Governance needs to come before scale. EY is building the oversight function now, after already scaling. The organizations that get ahead of this problem will build agent economics into their AI strategy before costs get away from them.

The emergence of agent economics as a formal discipline is a healthy sign. It means the industry is moving from “deploy AI and see what happens” toward treating AI as a managed asset with measurable returns. That’s a more serious, more sustainable approach — and the businesses that adopt it earliest will be better positioned to scale AI without the unexpected bills.

---

Enterprise DNA helps organisations build and manage AI agent workforces that deliver measurable business outcomes. Talk to us about building AI that actually pays for itself.