When AI takes over work people used to do, the cost moves out of the headcount budget and into the technology budget. Responsibility for the result often ends up with nobody in particular.
Only 14% of large US companies give accountability for AI-informed or AI-executed business decisions to the leader of the business unit where the work happens, according to KPMG’s Q2 2026 AI Quarterly Pulse Survey. The rest put it with a named C-suite executive (34%), the CEO or executive committee (32%), or a governance committee (10%).
Spending is shifting at the same time. Gartner’s 2026 CFO budget benchmarks show expected headcount growth falling from 6% in 2025 to 2% in 2026, while IT budgets are among those growing fastest. More of the work is moving to AI agents. But people sit in one budget, agents sit in another, and accountability sits above both. Nobody owns the trade-off between them. The teams that ship AI successfully are the ones that intentionally assign it.
Key findings
Accountability for AI-driven business decisions is split: 34% assign it to a named C-suite executive, 32% to the CEO or executive committee, 14% to a business unit leader and 10% to a governance committee, per KPMG.
Only 26% of large US companies say the operating costs of their AI systems are fully visible today.
CFOs expect headcount growth to fall from 6% in 2025 to 2% in 2026, and average HR budget growth to slow from 2.4% to 0.7%, per Gartner.
43% of leaders cite human oversight skills as a challenge in deploying AI agents, second only to data readiness at 58%.
Only 19% of companies are changing their mix of full-time, contract and gig workers in response to AI, per Deloitte.
Three budgets, one workflow
Take a single workflow, such as reconciling campaign delivery against invoices or monitoring bid performance across a set of accounts. The AI agent doing that work is paid for as software, usually from the IT budget, and increasingly on a usage-based model that bills by the token or the API call. The people who supervise it sit in a business unit’s headcount plan, which is administered by HR. The contract specialist who built the integration is in a procurement agreement. The outcome is something the business unit is measured on.
That means the most important workforce decision of the next few years is spread across three budgets. Should the team hire another analyst, run another agent, or bring in a specialist for six months? Those options can’t be compared directly, because no one sees all of their costs in one place. The budget side is just as fragmented.
Gartner’s numbers show which way the pressure points. When headcount growth drops from 6% to 2% while technology budgets climb, a manager under cost pressure has every reason to move work out of the headcount plan, where it’s scrutinized, and into the technology budget, where it often isn’t. The reduced headcount shows up in one budget as savings. The agent’s operating costs and the people needed to supervise it show up somewhere else, if they’re tracked at all.
Nobody can price the trade
KPMG found that only 26% of large companies say their AI operating costs are fully visible today. More than a third of leaders named AI cost and economic literacy, such as understanding token and inference pricing, as a challenge in deploying agents.
Agents also come with people costs that rarely appear in the business case. Someone has to evaluate their output, handle the exceptions they escalate, monitor them for drift and answer for their mistakes. KPMG found 43% of leaders cite human oversight skills as a challenge in deploying agents. That’s the same staffing gap Talent Observer covered in AI pilots stall on staffing piece, where pilots stall because nobody was hired for the production work.
Talent Observer also covered the HR version of this, wherein agents take over the routine work, and the people who remain become supervisors, which is a job few teams are staffed or trained to do. The same pattern applies in engineering, ad operations and analytics.
Accountability lands too high or too far away
Sixty-seven percent of leaders say their CEO actively owns AI as a strategic priority. That helps with funding and urgency. But a CEO can’t own the daily output of a reconciliation agent, and neither can a governance committee that meets once a month. A named C-suite executive such as the CIO usually owns the platform rather than the business result. That leaves the people closest to the work, who know when an output is wrong, with the least formal accountability for it.
Deloitte’s State of AI in the Enterprise 2026 found that companies where senior leaders actively shape AI governance get significantly more value than those that leave it to technical teams. It also found that some companies are merging their technology and people-leadership functions so that systems and workforce design change together. Both findings point to the same idea: senior leaders set the rules, and a named operating owner runs the work every day.
What ownership looks like in ad tech
Media and ad tech companies will run into this sooner than most. As covered in Hiring for agentic buying, an AI buying agent spends real money, adjusts pacing in real time and makes decisions with financial consequences in minutes, not quarters. If that agent overspends a campaign, someone has to be clearly responsible. A CIO who owns the platform or a committee that reviews policy isn’t that person. It should be the head of programmatic or ad operations who owns the campaign outcomes.
That owner’s budget should include everything the workflow costs: the agent’s operating costs, the engineers who evaluate and supervise it, any contract specialists building integrations, and the people who handle exceptions. The owner should also be judged on outcomes, not usage.
How to assign ownership
Teams that run human and agent work well tend to set it up like this.
Assign one owner per workflow, not per tool. The owner is accountable for the outcome, whether people, agents or a mix do the work.
Give the owner one budget for the work. People, agents, contractors and supervision costs go on the same line, so trade-offs can be compared directly.
Allocate AI costs to workflows. Metered costs that roll up to a single IT line can’t inform a staffing decision.
Staff oversight before go-live. Name who evaluates the agent, who handles its exceptions and who can shut it off.
Split the support roles clearly. IT runs the platform, security and cost metering. HR designs roles and keeps the full workforce map, including agents and contractors. The business owner decides the mix.
Revisit the mix regularly. Agent costs and capabilities change quickly, so the right balance of people, agents and contractors will shift.
The question of who owns the headcount is really about who owns the work. When people did all of the work, those were the same thing. Now they aren’t, and the companies that see the best returns from AI will be the ones that assign both to the same person.
Frequently asked questions
Who should own AI agents in a company? The business leader accountable for the workflow the agent supports. IT should own the platform and security, and HR should own workforce design, but daily accountability for an agent’s output belongs with the person who owns the outcome.
Who is accountable for AI decisions within a company? It varies. In KPMG’s Q2 2026 survey of large US companies, 34% assign accountability to a named C-suite executive, 32% to the CEO or executive committee, 14% to a business unit leader and 10% to a governance committee.
Should AI agents count as headcount? Not as employees, but they should be planned and budgeted alongside people. When agents sit in a software budget and people sit in a headcount plan, companies can’t compare the cost of hiring with the cost of automating.
Why are AI agent costs hard to track? Many agents are billed by usage, such as tokens or API calls, and those costs roll up to central IT budgets. Only 26% of large companies say their AI operating costs are fully visible today, per KPMG.
Do AI agents reduce headcount? They change it. CFOs expect headcount growth to fall to 2% in 2026, per Gartner, but agents need people to supervise them, evaluate their output and handle exceptions. Those roles have to be staffed for the agent to work.
What does HR’s role become when AI does more of the work? Designing roles that combine human and agent work, keeping a full map of employees, contractors and agents, and building the oversight skills agents need. HR should support the business owner’s staffing decisions, not make them alone.
What skills do teams need to manage AI agents? Evaluation, exception handling, cost literacy and judgment about when to override an agent. KPMG found 43% of leaders cite human oversight skills as a challenge in deploying agents.
Who should own an AI buying agent in ad tech? The head of programmatic or ad operations accountable for campaign outcomes, with a budget that covers the agent’s operating costs, the engineers who supervise it and any contract specialists who build its integrations.
