We Already Knew This Would Happen
Ignoring the past signals from adopting Cloud when implementing AI
Domain: Technology & AI | Arc 6: The Repeating Pattern | Theme: Cloud taught us this lesson. We ignored it.
A note on Jordan
Jordan is a fictional character. Their story is drawn from a combination of professional observation and personal proximity to real events. The experiences described are real. The person is not.
Jordan had been waiting for the answer to change, but it didn’t.
The meeting had started straightforwardly enough, a conversation with a senior leader at a UK government department about their cloud data usage and how things might be managed more effectively. The data estate had grown substantially. Costs were climbing. Teams across the department were accessing cloud services as though they were drawing from an inexhaustible supply — no budget line, no allocation model, no sense that every query, every storage call, every API interaction was a transaction with a cost attached.
Jordan asked what the plan was to bring it under control.
The answer was calm, unhurried, entirely serious. The bill would be paid. More budget would be allocated. That was the plan.
Not governance. Not a framework. Not a conversation about whether the consumption was producing value proportionate to its cost. Just more money, applied to an unmanaged system that would continue consuming at whatever rate its users decided, which is to say, at an accelerating one, because consumption-based technology without governance has only one direction of travel.
This was not a failure of intelligence. It was a failure of institutional memory. Because we had seen this before. We had, in fact, already learned this lesson. We had simply chosen not to apply it.
Cast your mind back to the first wave of cloud adoption. The pitch was familiar: move to the cloud, pay only for what you use, reduce your capital expenditure, increase your flexibility. All of that was true, in the narrow sense that each individual statement was accurate. What was not said, or not said loudly enough, was that consumption-based technology in the hands of an ungoverned organisation does not reduce costs. It redistributes them, obscures them, and then grows them.
Teams who had never had to think about infrastructure costs suddenly had the ability to spin up compute and storage at will. The people making those decisions were rarely the people receiving the bills. There was no common framework for deciding what consumption was justified, no mechanism for attributing cost to the teams generating it, no function responsible for ensuring the organisation was getting value for what it was spending. The Cloud Business Office — the governing function that AWS, Microsoft, and every serious cloud governance framework had described in detail — existed in policy documents and almost nowhere else.
The result was predictable, and it was predicted. Cloud bills grew faster than cloud value in organisations without governance. The ones that got it right built a Cloud Business Office with real authority: a cross-functional team responsible for consumption standards, cost visibility, budgeting models, and the mandate to say no when the answer to why are we doing this? turned out to be because we can.
The ones that got it wrong paid the bill and allocated more budget.
Now look at what is happening with AI.
The pattern is identical, and we are watching it repeat in real time.
AI platforms, particularly the large language model platforms being deployed at enterprise scale, are consumption-based. Every query costs money. Every document processed, every image analysed, every automated workflow triggered is a transaction. The pricing models are designed to be frictionless at the point of use, which is another way of saying they are designed to obscure the true cost until the invoice arrives.
And just as with cloud, the people making consumption decisions are rarely the people accountable for the budget. A team discovers that they can use an AI platform to draft reports, summarise meetings, analyse documents. They use it. A lot. Why wouldn’t they? It feels free, or close enough to free that the friction is invisible. Nobody told them it wasn’t. Nobody built the structure that would make the cost visible at the moment of consumption.
This is not a technology problem. It is a governance problem, and it has a known solution: the AI Business Office.
The Cloud Business Office model, recognised across the major cloud frameworks, had four core responsibilities.
It aligned technology consumption with business outcomes.
It created cost visibility and attribution making teams accountable for what they used.
It set standards for what consumption was justified and what was not.
And it had the authority to enforce those standards, not merely to recommend them.
The AI Business Office is the same function applied to the next wave. Its responsibilities are not materially different, but the stakes are higher in at least three respects.
First, AI consumption is less predictable than cloud consumption. A storage allocation scales in a relatively linear way. AI usage, once users discover what it can do, scales exponentially because the more useful something is, the more people use it, and AI tools tend to surface new use cases faster than governance can respond to them.
Second, the inputs to AI systems carry risks that cloud storage inputs did not. A file uploaded to cloud storage is still your file, in your control, sitting in a location you have selected. A document entered into an AI model, particularly a third-party hosted model, may or may not stay within your control, depending on the platform’s data retention policies, the terms you agreed to, and the security architecture of the service. Most users have no idea which of those applies to the documents they are processing. The AI Business Office is the function that finds out, sets policy, and ensures people follow it.
Third, the value question is harder. Cloud usage is relatively easy to trace: this team stored this much data, ran this many compute hours, generated this cost. AI value is murkier, a report drafted faster, a meeting summarised instead of attended, a decision supported rather than made. That murkiness is not an argument against using AI. It is an argument for having a function that insists on measuring value rigorously before consumption is approved at scale.
There is a broader version of this problem that most organisations have not yet considered.
When everyone can build an application, and they can, because the vendor who built one in two hours without being a developer was telling the truth, the question of what is being built across the organisation becomes urgent. In the cloud era, ungoverned teams created orphaned infrastructure: compute instances nobody owned, storage buckets nobody could account for, integrations nobody had documented. The technical debt from that period is still being paid off in many organisations.
AI-assisted development compounds that problem. The applications being generated are more complex, less visible, and more embedded in business processes than a forgotten storage bucket. They carry assumptions about data, about outputs, about reliability that the person who clicked the prompts did not know they were making. The AI Business Office is not a brake on this, it is the framework that allows it to happen responsibly, with visibility, with standards, with a clear line of accountability when something breaks.
Which it will. The question is whether anyone knows whose job it is to fix it.
The Cloud Business Office arrived, in most organisations, late. It was built after the bills became uncomfortable, after the auditors asked questions, after someone in finance noticed that the cloud was not, in fact, saving money. The conversation that preceded its creation was almost always some version of: we knew this was coming, we just didn’t build the structure in time.
That conversation is going to happen again about AI. The only question is whether your organisation has it now, before the bill arrives or after, when the answer is still to pay it and allocate more budget.
The lesson was already learned. The solution is already known. The Cloud Business Office was not a perfect model, but it was a real one, built from real experience of what happens when consumption-based technology runs without governance.
We do not need to invent anything new. We need to apply what we already know.
Four questions to sit with:
Does your organisation have a function? not a policy document, but an actual team with authority responsible for AI consumption, AI cost visibility, and AI standards? If so, who leads it? If not, who should?
Do your teams know what it costs every time they use the AI platforms they have access to? Not approximately, exactly. And do they know who is accountable for that cost?
When a team member builds something with AI assistance — an application, a workflow, an automated process — is there a review process that understands what they have built, what it depends on, and what happens when it fails?
And the question beneath all of the others: the Cloud Business Office was the lesson we were supposed to have learned. Are you applying it, or are you waiting for a conversation in which someone tells you the bill will just be paid?
You’re reading The Next Evolution by Neil Catton, articles that explore the human world and the intersection of technology, they try and ask difficult questions - not to scare - but to inform. If someone forwarded this to you, you can subscribe free at neilcatton.substack.com.
Neil Catton is the author of The Next Evolution, The Cognitive Crucible and The Shadow System - available on Amazon, and writes at the intersection of technology, ethics, and human purpose.