Council Post: Why Agentic AI Is About To Give CIOs Their Cloud Bill Moment

Andi Mann is Chief Product and Technology Officer at Apica, a leader in telemetry pipelines for observability platforms.

getty

There is a moment every CIO remembers: when the cloud bill arrived.

Not the projected cost. The actual bill showed what happened when every department and developer started spinning up compute and storage without anyone tracking aggregate spend. The one that reflected months of "just take care of it" authorization from leadership, who assumed IT controlled something IT had never been asked to control.

Some CIOs lost their jobs over that bill. Others spent years unwinding architectural decisions made in the gold rush of cloud adoption. The organizations that thrived saw it coming.

I am telling you now: We are about to see this again. The bill is being written, and now, it is denominated in tokens.

From MIPS To Tokens: Everything Old Is New Again

I have spent more than 30 years in enterprise IT, mainframe, midrange, client-server and cloud. Each shift followed the same pattern: A powerful new capability arrived, business units adopted it faster than finance could model it, and IT was left holding an unapproved cost structure.

We used to call the unit of compute consumption MIPS, millions of instructions per second, and the CIO was perpetually arguing with the CFO about who authorized them. The business units were running them up; IT was carrying the bill. Sound familiar?

Today, the unit is tokens. Every agentic AI workflow, every autonomous agent querying a knowledge base, every LLM call embedded in a business process, every multistep reasoning chain, consumes tokens. Costs are accumulating across business-unit budgets, shadow IT projects and developer experiments at a scale most CIOs have yet to quantify.

New research from Omdia, commissioned by Apica and surveying 300 enterprise IT decision-makers, found that 59% of enterprises have already terminated or delayed an agentic AI deployment because observability costs have become unmanageable. This is already happening. Enterprises pausing projects now notice early on. The ones who haven't are building a much larger bill.

The Authorization Gap

Here is the structural issue: The CIO is not deploying agentic AI agents. The business units are. Product teams are. Individual developers with API keys and credit cards are.

No one walked into a CIO's office and said: "I'd like to authorize 50 million tokens a month for this new customer service workflow." They said: "We're automating customer service with AI."

An EVP at a major e-commerce company told me he started measuring token consumption as a core performance metric, not by design, but because he was surprised by the cost and had to build accountability structures retroactively. That is the institutional scar tissue that forms when governance lags adoption.

The Omdia research validates this at scale: Enterprises face observability costs for agentic AI that now exceed the cost of the AI infrastructure itself. The cost of watching agentic AI run is outpacing the cost of running it.

This Is Not Just A Budget Problem

CIOs who frame this as purely a cost issue are missing the deeper architectural problem.

Agentic AI generates data differently from traditional applications. Agents operate in loops, make sequential calls, branch, backtrack and spawn subagents. A single customer interaction can generate telemetry spanning dozens of micro-decisions, tool calls and external API requests, all of which occur faster than any human operator can observe in real time.

Most enterprises are trying to monitor this with observability infrastructure built for a fundamentally different world, discrete transactions, batch processes and human-speed workflows. Sending agentic telemetry into a platform built for microservices monitoring is like using a sedan's fuel gauge to manage a fleet of Formula One cars. The instrument exists. It just was not built for this, and no amount of dashboard-adding will fix a pipeline architecture problem.

What CIOs Should Do Now

Don't wait for the bill to arrive.

The Omdia data shows that 68% of enterprises plan to evaluate changes to their observability solutions within the next six months, your window to get ahead of the cost curve rather than respond to it. Here is where to start:

Get visibility into token consumption now. Work with your AI and engineering teams to instrument token usage across every agentic workflow: formal ones, shadow projects and developer experiments. You cannot manage what you cannot measure.

Treat compliance as the baseline, not the finish line. PII, PHI, IP protection, table stakes, not aspirations. If data governance controls are not in place before your agents start operating autonomously, you're not running an agentic AI program; you're running an uncontrolled experiment.

Audit your telemetry architecture against the new workload. Ask your observability team one question: "Was this platform designed for agentic AI data volumes?" If the answer is no, or if they hesitate, you have work to do.

Build cost attribution before the business units scale. The window is now, while agentic deployments are still relatively contained. Once 10 business units have autonomous agents running at scale, retroactive attribution becomes politically and technically nightmarish. Set the model while you can.

The Leaders Who Get This Right Will Define The Next Decade

Every major platform shift produces a new cohort of IT leaders: those who dove in early, built expertise when it was hard and became reference points for their peers. We saw it with cloud and DevOps. The CIOs who navigated the cloud cost crisis became the most sought-after executives in enterprise technology.

Agentic AI is in that moment for this generation. The organizations that get observability and cost governance right will have better data, better predictions and better ability to scale AI confidently. The ones that wait are going to spend the next two years unwinding decisions that were never authorized.

The bill is being written. The question is whether you'll be in a position to read it before it arrives or whether you'll be the one explaining it to your board.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?