AI Agents Are the New Cloud Spend Problem
There’s a familiar narrative forming around AI agents. They’re going to replace work, drive efficiency, and reduce cost. It’s the same promise we heard in the early days of the cloud. And just like then, the conversation is moving fast—faster than most organizations can structure around it.
But what’s starting to show up tells a different story. Without controls, AI agents can cost more than employees. Not because the technology is flawed, but because the system around it hasn’t caught up.
This isn’t new.
It’s cloud all over again.
If you were part of the first wave of cloud adoption, you remember what happened. Teams moved quickly. Resources were spun up instantly. There was little tagging, limited ownership, and almost no accountability.
Everything worked exactly as expected
until the bill showed up.
AI agents follow the same pattern, but they move faster and operate with more autonomy. They don’t just consume infrastructure. They make decisions, trigger actions, and execute workflows.
And without guardrails, they don’t just run.
They scale spend.
What’s different this time is the type of cost we’re dealing with. Cloud introduced variable infrastructure cost.
AI agents introduce variable decision cost.
You’re no longer paying just for compute. You’re paying for every step an agent takes—every API call, every iteration, every loop, every layer of reasoning.
And unlike traditional systems, those steps aren’t always predictable.
This is where things start to break.
I’ve written before that FinOps isn’t about cutting cost. It’s about visibility, ownership, and accountability around consumption.
The same principles apply here.
If AI agents are deployed without defined cost boundaries, ownership tied to business units, and clear policies for execution, then what you have isn’t innovation.
You have unbounded consumption with autonomy.
And that’s significantly harder to control.
One of the biggest gaps I’m seeing is that we’re deploying AI agents without designing the cost architecture around them.
In cloud, we learned to think in tagging strategies, cost centers, budgets, alerts, and workload-level accountability.
AI agents need the same discipline.
Every agent should have a defined purpose, a cost boundary, an owner, and a measurable outcome.
Otherwise, you’re not deploying capability
you’re introducing financial risk.
This is where the operating model gap shows up.
Technology adoption moves fast.
Operating models lag behind.
That gap is where problems live.
We’ve seen it with cloud. We’ve seen it with security.
Now we’re seeing it again with AI.
AI agents aren’t just tools.
They operate more like digital workers.
And yet, we’re deploying them without the structure we’d apply to actual employees.
No job description.
No budget.
No accountability.
That doesn’t scale.
I keep coming back to the idea of a North Star.
With cloud, the goal was clear, move, migrate, scale. We reached it, and then spent years retrofitting governance, cost control, and security.
Now we’re doing the same thing with AI.
So the question becomes:
Is AI our next North Star… or are we repeating the same cycle?
Because if we don’t define it clearly, we risk prioritizing speed over structure.
And that always shows up somewhere
usually in cost, risk, or both.
AI agents aren’t the problem.
Uncontrolled systems are.
Cloud taught us that consumption without accountability leads to waste.
AI takes that same dynamic
and adds autonomy.
Technology is accelerating.
But without the right operating model, that acceleration creates instability.
You don’t get efficiency.
You get chaos—with a billing line attached to it.
The organizations that get this right won’t be the ones deploying the most AI agents.
They’ll be the ones that apply cloud discipline to AI early.
They’ll define ownership.
They’ll build cost architecture.
They’ll establish guardrails before scale.
Because if they don’t—
the outcome is predictable.
It won’t show up immediately.
But eventually… the bill always does.
This is part of my cloud series on cost, operating models, and modernization