Most companies track AI as a software line item. But AI is increasingly doing the work, not just supporting it, and that means the real question isn't "how much are we spending on AI?" but "what outcomes is it producing, and who owns them?"Nobody in the room could answer. Not because the team was careless, but because nothing they had built over the past twenty years was designed to answer it.That's the gap I want to talk about. In this article I'll explain why AI breaks the way companies traditionally manage technology, why rising AI costs are a symptom rather than the disease, and the five-part governance model I've seen actually work.Every Tech Wave Creates a New Management DisciplineEvery major wave of enterprise technology changes not just how companies work, but what they have to manage. Cloud turned compute into an on-demand resource, and we got cloud economics and FinOps. SaaS gave us sprawling portfolios of subscriptions, and we got software procurement, access management, and vendor governance.Generative AI is the next wave. And most companies haven't built the discipline for it yet.The scale makes this urgent. Gartner expects global AI spending to reach $2.59 trillion in 2026, up 47% year over year. Meanwhile, McKinsey's State of AI 2025 found that 88% of organizations now use AI in at least one function, yet only 39% can attribute any EBIT impact to it, and for most of those, it's under 5%.That's a lot of money flowing into something most companies can't measure.The Real Problem Isn't Rising AI CostsMost of the enterprise AI conversation right now is about cost, and there's reason for concern. Gartner predicts that through 2028, at least half of GenAI projects will overrun their budgets because of poor architectural choices and a lack of operational know-how.But I'd argue cost is a symptom. The real problem is that most companies have no visibility into what they're actually paying for.Try answering these about your own organization:Where is AI actually being used, including tools teams adopted without asking?Which workflows are producing measurable value?Which teams consume the most AI resources, and are any of them duplicating each other?Which automation can you point to and say, with confidence, "that made us money"?Which models are the right fit for which tasks?In my experience, most leadership teams can't answer these with any precision. Finance sees AI spend spread across model usage, infrastructure, data services, integration, and human review, but can't tie it to outcomes. Operations tracks adoption numbers without knowing which workflows matter. IT manages a growing web of models, copilots, and agents without a clear picture of how they overlap or what dependencies they've quietly created.The gap isn't spending control. It's visibility, evaluation, and attribution.Why Cheaper Tokens Won't Save YouHere's the part that surprises a lot of executives. The unit price of AI is falling fast. Gartner forecasts that by 2030, running inference on a trillion-parameter model will cost providers over 90% less than in 2025.And yet, the same firm expects inference cost per agentic workflow to more than quintuple by 2028, because agents use roughly 5 to 30 times more tokens per task than a standard chatbot.Cheaper tokens, more expensive outcomes. If you're only watching the price per token, you're watching the wrong number.AI Adds a New Layer to Operational Cost: Decision-Making CapacityFor decades, companies have built increasingly precise systems to measure what it takes to run the business. First people and labor capacity. Then software. Then infrastructure. Cloud made consumption dynamic: it was no longer enough to know what technology you owned; you had to know how much you were using.AI adds a new layer on top of all that: decision-making capacity.AI isn't just supporting work anymore. Increasingly, it does parts of the work itself. It analyzes contracts, prioritizes sales leads, catches anomalies, answers customers, writes code, and routes requests. As agents get more capable, they coordinate these tasks with less and less human involvement.Once that's true, treating AI as a software line item stops making sense. The right question becomes: what operational capacity are we buying, where is it applied, and what outcomes does it produce?Why Traditional Governance Breaks DownTraditional software governance was built for stable objects. A license is a license; an app does roughly the same thing every day.AI doesn't work that way. Its behavior shifts with the model, the prompt, the data, the context, the connected tools, and the surrounding workflow. The same business task might be handled by different models at very different costs and quality levels. An agent might call other systems, consume extra resources, or make decisions that affect customers and employees, without anyone approving that specific chain of events.So AI governance has to move from "who owns the software" to "who owns the outcome."Consider customer service. The question isn't just "did we approve this model?" It's:Does this workflow actually resolve issues correctly?Is it escalating the right cases?What does each successful resolution cost?Who is on the hook when performance slips?That last cost question matters more than it seems. Gartner predicts that by 2030, GenAI cost per resolution in customer service will exceed $3, more than many offshore human agents. A company that only tracks "AI adoption in support" will never see that coming. A company that tracks cost per resolution will.A Governance Framework That Actually FitsModels will keep changing, and organizations will keep using many at once. So governance can't be built around controlling individual models. It has to focus on the systems AI operates within. In my view, that comes down to five principles.1. Visibility. Maintain a current map of where AI is used across applications, workflows, teams, and agents. You can't govern what you can't see.2. Attribution. Connect usage to outcomes. A million model calls tells you nothing on its own. What did those calls achieve?3. Optimization. Not every task needs your most powerful, most expensive model. Route work across models based on cost, speed, risk, and the quality the task actually requires. Gartner makes the same point: routine, high-frequency tasks should go to smaller, domain-specific models.4. Accountability. Every AI-driven workflow needs a named owner, responsible not just for deploying it but for how it performs over time.5. Continuity. AI systems move too fast for annual reviews and static approvals. Cost, behavior, quality, and dependencies can shift week to week. Monitoring has to keep pace.This Isn't Just an IT ProblemAI governance can't sit with IT alone. It's becoming an executive discipline across finance, operations, technology, and business leadership. McKinsey's research points the same way: the companies seeing the most value from AI are the ones that redesign workflows and put senior leadership in charge of AI governance.Companies that keep treating AI like a SaaS subscription will end up with fragmented adoption, duplicated capabilities, rising costs, and nobody clearly accountable. Companies that treat it as a core operational resource will know where their intelligence is applied, what it costs, what value it creates, and who owns the result.Three Questions to Ask Your Team on MondayIf you want to know where your organization stands, start here:Can we list every AI-powered workflow we run, and name one owner for each?For our three biggest AI use cases, do we know the cost per successful outcome, not just total spend?When did we last check whether a cheaper model could do the same job just as well?If the answers are "no," "no," and "never," you're not alone. But that's the conversation your leadership team needs to have next.Irina Shymko is CEO at Langate EMEA, where she helps enterprises across Europe and the Americas turn emerging technology into measurable business outcomes.