A new wave of financial pain is hitting enterprise AI deployments. Axios reported this week that “AI sticker shock” is sweeping corporate America, with organisations discovering their AI budgets were built on assumptions that no longer hold.

The cause is agentic AI. As businesses move from simple chatbots to AI agent workflows — systems that reason, plan, and execute multi-step tasks — token consumption is exploding in ways that finance teams never modeled.

The Numbers Are Alarming

The pattern is showing up across industries. One healthcare enterprise consumed one trillion tokens over six months, generating more than $6 million in unplanned costs. Uber’s CTO publicly acknowledged the company burned through its entire 2026 AI coding tools budget in just four months. A CloudZero report found that 49 percent of organisations say they cannot confidently calculate a return on their AI investment, largely because AI spending is scattered across cloud providers, GPU services, API vendors, and SaaS platforms with no two billing formats alike.

Gartner estimates that 65 percent of enterprises deploying generative AI will exceed their budget projections through 2026, driven by token consumption patterns that only emerge after widespread employee adoption begins.

The core problem: finance teams modeled AI like traditional software, as a fixed seat license or predictable monthly fee. Agentic systems do not work that way.

Why Agents Cost So Much More

The shift from chatbots to agents is not a minor upgrade. Agentic workflows require five to thirty times more processing per task than the simpler prompt-and-response tools that most cost projections were built around. Each time an agent plans, retrieves information, calls a tool, checks its work, and iterates, it consumes tokens. A task that costs cents in a basic chatbot interaction can cost dollars or more in an autonomous agent workflow.

Goldman Sachs quantified the scale of this shift in a recent research report: AI agents could drive a 24-fold surge in global token demand by 2030. Enterprise agents are projected to account for over 70 percent of all token usage by 2040. A programming agent could consume 7 million tokens per day, while a data entry agent might consume 25 million.

That is not a rounding error. That is a fundamental rethink of how AI is priced, governed, and deployed.

The Billing Change Nobody Is Ready For

Adding urgency to the story: GitHub Copilot shifted to usage-based billing on June 1, with token-linked AI credits priced at one cent each. Agentic Copilot workflows can consume 40,000 tokens before a developer has had their morning coffee. The move signals a broader industry shift away from flat-rate AI subscriptions toward consumption pricing, a model that rewards efficient use but punishes organisations that have not built governance around their AI deployments.

Microsoft has already begun revoking internal developer access to certain AI coding tools, citing cost control. When one of the world’s largest technology companies is throttling its own AI usage over budget concerns, it is worth paying attention.

What This Means for Business

The sticker shock story is not a reason to avoid AI. It is a reason to approach AI deployment with the same rigour you would bring to any major capital investment.

Most organisations are still treating AI like a productivity tool they can enable and forget. Agentic AI requires active management: monitoring token consumption, setting usage policies, identifying which workflows genuinely justify the cost, and building ROI frameworks before the bill arrives.

A few practices that separate disciplined AI adopters from the rest:

• Start with a cost model, not a use case. Before deploying any agent workflow, estimate token consumption under realistic usage assumptions. Build in a buffer.
• Govern from day one. Know who in your organisation is using AI, on what tasks, and at what cost. Consumption monitoring is not optional at scale.
• Measure outcomes, not activity. Track what the AI actually achieved: time saved, revenue generated, errors avoided, not just how many queries were run.
• Right-size the model. Not every task needs the most powerful model. Routing routine queries to smaller, cheaper models while reserving reasoning-heavy agents for complex tasks is where significant cost savings are found.

The Case for Advisory Before Implementation

What the sticker shock stories share is a common failure mode: organisations deployed AI before they understood the operational economics of AI. The technology moved fast and the governance did not keep up.

This is precisely the gap that strategic AI advisory is designed to close. Getting independent expert input on your AI roadmap before you have committed to infrastructure, vendor contracts, and agent architectures is not a luxury. Given the cost trajectories described above, it may be the highest-ROI investment a business makes in its AI journey.

The age of “try it and see what it costs” is over. Enterprise AI is now a significant line item that requires the same financial discipline as any other major operational investment.

---

Enterprise DNA works with business leaders navigating AI strategy, from evaluating vendors to designing agent workflows that actually deliver measurable returns. Learn more about Omni Advisory.