Ask "who owns AI spend in a company" and most answers point at a department: engineering, IT, finance. That is the wrong unit. Ownership only means something at the level of a specific tool, a specific agent, a specific budget line, with one name attached. Departments do not get surprised by an invoice. People do.
Only 34% of C-suite executives in Pearl Meyer's Q2 2026 survey said it was consistently clear which executive or team makes the calls on AI. That was the lowest score of any group surveyed. Board members felt more confident, at 53%. Senior managers below the C-suite felt more confident still, at 57%. Read that order again. The people closest to actually implementing AI are the most convinced someone else has it handled. The people at the top, who would presumably be that someone else, are the least sure it is them.
"Ambition for AI outcomes is currently outpacing the leadership structure needed to deliver on them," the Pearl Meyer study put it. "Additional investment without clear ownership will only widen that gap."
That is the decision-making version of the problem. The spending version looks almost identical. Harness surveyed 700 engineering leaders in its 2026 State of AI in FinOps report and found that 52% say there is no clear, dedicated owner of AI cost at their organization, with responsibility split across engineering, FinOps, finance, and IT. Split four ways is another way of saying owned by nobody. The same report found organizations estimate 26% of all AI spend is wasted, with no owner positioned to explain where it went.
Sharp insight: when a decision has four partial owners, it does not get made four times as carefully. It gets made once, by whoever moved first, and nobody else finds out until the invoice does the telling.
Why the gap is not an enterprise problem
It is tempting to read both of these reports as a big-company story: too many committees, too many layers, too much distance between the board and the work. That reading misses what is actually happening at a 200-person company, which is usually worse, just quieter.
A Fortune 500 company at least has a FinOps team, a procurement process, and someone whose job title includes the word "governance." A 200-person company usually has none of that. What it has instead is a founder who approved the first three AI tools personally, an engineering lead who has since added a dozen more without asking anyone, and a finance function that finds out a new agent exists when the credit card statement arrives.
Harish Doddala, VP of Product at Harness, put it plainly: the pattern shows up "the same way whether we're talking to a 200-person startup or a Fortune 500 company." The difference is not whether the ownership gap exists. It is that a Fortune 500 company has people whose entire job is to eventually notice it. A 200-person company does not.
Three ownership models that look fine on paper and fail in practice
Most companies do have something written down about AI ownership. It just does not survive contact with an actual invoice.
"Whoever bought it owns it." This sounds reasonable until the person who bought it changes teams, or leaves, or simply stops thinking about the tool once it works. Ownership by original purchase decays fast, and nothing replaces it when it does.
"IT owns it." IT can own access and security. It rarely owns whether a tool is worth what it costs, because that requires knowing what the tool was supposed to accomplish in the first place, which is a business question, not a systems one.
"Everyone owns it together." This is the version Harness measured directly, responsibility split across engineering, FinOps, finance, and IT, and it is functionally identical to nobody owning it. Shared ownership without a single accountable name is a diffusion strategy, not a governance one.
All three feel like answers when someone asks the question out loud. None of them survive the actual test, which is: pull up one tool right now and name the person who would explain its cost to a board member in the next five minutes.
What ownership actually means, broken into parts
Part of why this stays unsolved is that "who owns AI spend" sounds like one question and is actually three.
Who decides what gets bought. This is usually whoever has the loudest need in the moment, an engineer who wants a coding assistant, a support lead who wants an agent for tickets. Nobody is wrong to want the tool. The problem is that "I needed it" and "I own it" get treated as the same sentence.
Who is accountable for what it costs. This is supposed to be finance, except finance usually cannot see the number until it already happened, and by then accountability has quietly turned into archaeology.
Who reviews whether it is still worth it. This is the part that almost never has an owner anywhere, at any company size, because nobody's job description includes "revisit a decision that already shipped."
A single person or team rarely holds all three. That is not necessarily wrong. What is wrong is when nobody holds any of them, which is closer to the default state most companies are actually operating in right now.
Sharp insight: ownership is not a name on a spreadsheet. It is the answer to a specific question, asked cold, about a specific tool: if this agent's usage tripled next month, who would be the first person to notice, and would they know why it happened.
What ownership actually looks like in a 200-person company
Strip away the org chart language and real ownership at this size comes down to four habits, not four departments.
One name per tool, not one team. "Engineering owns AI tools" is not an answer, it is a deflection with extra steps. The question that actually produces accountability is narrower: who owns the Cursor seats, specifically, by name. Who owns the support agent. Who owns the internal chatbot nobody remembers approving. If the honest answer is "nobody in particular," that is the finding, not a footnote.
A budget set before deployment, not discovered after. Most AI spend at this size is uncapped by default, not because anyone chose that, but because nobody chose anything. Setting a number before a tool goes live, even a rough one, turns the first month of usage into a signal instead of a surprise.
A standing review date, not a one-time approval. The tools that get evaluated once, at purchase, and never again are exactly the ones Harness found accounting for a quarter of all wasted spend. A recurring 15-minute look, monthly is enough, does more than most governance policies twice its length.
A shared view across finance and engineering, not two separate ones. The Harness report found engineering and platform teams hold 35% of the influence over cost-driving decisions like model choice and retry logic, while platform teams are held accountable for 30% of the cost and FinOps for 27%. When the people making the decision and the people holding the bill are different people looking at different dashboards, neither one is actually watching.
None of these four things require a platform team or a governance function. They require someone deciding, once, that "somebody probably has this" is not the same as somebody having it.
Picture how this plays out at an actual 200-person company. An engineering lead signs up for a coding assistant in February to unblock a sprint. By June, the whole engineering org is on it, usage has crept up with headcount, and the invoice has roughly quadrupled. Nobody lied, nobody overspent on purpose. The tool simply never had an owner past the first signup, so nobody was positioned to ask, at month three or month four, whether the plan still matched the usage. That question costs five minutes to ask and nobody was ever assigned to ask it.
Agents just made this worse, on a shorter clock
Everything above was already true before agents entered the picture. Agents do not fix the ownership gap. They compress the time you have to notice it.
Haseeb Qureshi, general partner at Dragonfly, described the shift bluntly after backing Sapiom's Series A this month: "Agents are becoming employees with no manager and no budget, and increasingly, the CTO is the one acting as CFO, allocating real money with no visibility into where it goes." An agent does not wait for a monthly review to run up cost the way a subscription does. It can burn a week's budget on a Tuesday afternoon, retrying a failed step in a loop nobody is watching, and the first person to find out is whoever opens the invoice.
A human hire without a manager gets noticed within a week. An agent without an owner can run for months.
