Who in the C-Suite Should Own AI?

In January 2026, the CEO of a Fortune 500 insurance company convened his senior team to discuss ownership of the company’s AI initiatives. The CIO felt the answer was obvious: agentic AI systems roll up to her. The COO countered that an agentic workforce is the definition of ops. The CFO noted that an AI system was already making underwriting decisions with direct P&L impact. The Chief Risk Officer observed that autonomous decision-making systems are a major risk exposure. The CHRO saw AI agents as functionally equivalent to workers, which partly fall under his remit. The Chief Data Officer reminded everyone that the whole kit and caboodle depended on the right permissions and data access, and this was her domain. The meeting ended without resolution. A few months later, the company drafted a job requisition for a CAIO-type role, joining peers such as Dell, Pfizer, PwC, UBS, and Expedia. Variations of this are playing out in C-suites everywhere. Agentic AI—goal-directed, contextually aware, autonomous systems able to call other software tools, make complex decisions, and take actions—is creating a tussle for turf at the top. The question of who controls AI is the critical org-chart issue at the dawn of the AI era, and it will influence a company’s strategy, investment levels, and the distribution of power and influence among leaders. This jostling for influence isn’t a new problem. One of the most insightful frameworks for understanding it was developed by sociologist Andrew Abbott in his landmark 1988 book, The System of Professions. Abbott showed that professional groups are locked in a perpetual contest over who controls which domains of work, and that major technological or social disruptions are the moments when those boundaries get redrawn. Applied to the C-suite, his theory reveals why the current scramble over AI will become heated, and it points to a practical way for leaders to move past the turf war and toward an organizational structure that actually works. Jurisdictional Competition Abbott argued that the professions exist in an interdependent system, perpetually competing with one another for control over domains of work. A profession will claim certain kinds of work that become the jurisdiction of its members. Think of it as their professional turf. The thing is, these jurisdictions aren’t permanently settled. They are won, defended, and lost in ongoing contests at the interstices between professions. Nurse practitioners and physician assistants now perform primary care tasks once reserved for general practitioners. CPAs and tax lawyers spar over who owns complex tax planning, just as attorneys and financial advisors both work on estate planning. Social workers compete with psychologists, who compete with psychiatrists, who competed with neurologists to treat mental health conditions. Jurisdictional competition has long existed inside organizations as well. The major corporate functions—supply chain, operations, R&D, information systems, product management, HR, finance, marketing, sales—are components of a system of interdependent work just like the professions, and each function claims certain decision rights as its own. In established organizations in ordinary times, the division of labor between corporate functions is stable. In periods of massive change, though, stability gives way to ferment. We’ve entered such a time now. It’s the organizational equivalent to the fall of an empire: stakeholders sense the corporate map will be redrawn, and each function presents reasons why territory should belong to it. The boundaries between functional leaders that had hardened over the years are suddenly becoming negotiable. Agentic AI is creating entirely new areas of work with no clear owner, but it is also shifting decisions that just a year ago were anchored to specific functional turf. Jurisdictional competition is back, and the contest has only just begun. Why Agentic AI Is Different In some sense, there’s nothing new here; previous shifts in technology triggered confined debates about turf. The arrival of ERP systems, for instance, created tension between business process owners and IT. Eventually the “control plane”—the right to set the rules, permissions, and orchestration logic—often settled under the CIO, with heavy input from the relevant domain owner. The Cloud initially seemed like it would decentralize compute, removing the need for central control. By shifting the high capex to buy hardware to the (initially) low opex of renting it, a business unit leader could swipe a credit card and spin up some software. However, that freedom proved to be temporary. As cloud spend ballooned and security risks and compatibility issues surfaced, the CIO reimposed central control over the shadow IT orgs that popped up in business units. Agentic AI will create far more radical change. Returning to the insurance company, an AI agent that autonomously processes claims truly is a technology system (CIO), an operational workflow (COO), a risk management concern (CRO), a financial decision-maker with P&L impact (CFO), a novel form of labor (CHRO), and a user of data (CDO). The agent’s decisions and actions span functional jurisdictions and it is core across most or all of them. Unlike SaaS platforms like Salesforce, Workday, and NetSuite, agentic AI collapses the distinction between “the tool” and “the work.” As Silicon Valley pundits are now fond of saying, with agentic AI, the software is the service. In short, the agents are the underwriting team, or at least they author the first draft of its judgment. Therefore, whoever governs the agents—their roles, permissions, guardrails, evaluation, escalation, and budgets—will increasingly control what and how the company decides. The Contestants and Their Claims Abbott observed that professions win jurisdictional contests by framing problems in terms that make their expertise indispensable. That’s happening in today’s C-suite. The CIO’s claim rests on infrastructure. AI agents must be built, integrated into existing systems, secured against novel attack surfaces, and maintained at scale. Creating and running these systems requires technical proficiency and stewardship. The COO sees agentic AI as a component of the core ops of the firm. Agents will live inside workflows, and workflows are operations. The COO’s claim is that agents exist to execute business processes, and optimizing processes is why the office of the COO was created. The CHRO’s claim is conceptually provocative. Agents are functionally a new category of workers. They need to be trained on company-specific knowledge, enculturated, supervised during deployment, evaluated on performance, and paired with human colleagues in ways that preserve morale, culture, and institutional knowledge. And they will complicate many aspects of managing the human workforce, creating landmines for HR to navigate. The CFO’s claim rests on economic accountability. Agents will eventually change approvals, pricing, asset utilization, and company spend in ways that move the numbers. Agents that are authorized to commit the firm to a payout, a contract, a credit decision, or a refund require financial accountability. Risk and legal leaders see autonomous systems as a novel form of failure, liability, and regulatory exposure. As we’ve already seen, when AI agents deny a loan or job application, reject an insurance claim, or are involved in an automobile accident, litigation and regulatory scrutiny may result. Finally, the CDO notes that every other functional claim is downstream of the data. Because agents reason over information, their outputs are only as sound as their inputs. Who has access to what data, under what governance rules, will determine whether an agent works at all, much less whether it works securely. From “Who Owns AI?” to “Who Owns Which Decisions?” Each of these claims is sensible. When multiple leaders have legitimate claims to the same jurisdiction, organizations typically default to one of four responses: Hand it to the most capable leader Give it to the squeakiest wheel Ship it off to a committee Pass it to whoever has existing budget However, none of these will work for agentic AI. The stakes are too high for politics, the technology is moving too fast for non-expert committees with competing interests to keep pace, and the scope is way too broad for any single function’s existing desires or capabilities. We need a different approach. That approach should begin by shifting the question from “who owns AI?” to “who owns which AI-related decisions?” The first question frames the issue as zero sum and invites a land grab, but the second decomposes the problem into specific decisions and assigns each to the person best positioned to make it. Think about all the choices governing an AI agent requires. Someone decides what the agent is trying to accomplish and how success is measured. Someone has to ensure it is built securely and integrated into existing systems. Someone has to set the rules about what data the agent can access and how it reasons over that data. Someone has to determine what level of autonomous action is acceptable, and what requires human approval. Someone has to figure out how agents and people work together, and how introducing agents changes the work people do. And someone has to control the upfront investment and inference budget and measure whether the investment is paying off. These are different decisions requiring different expertise. The COO or P&L owner is the natural authority on what agents should accomplish; the CIO on how to build and run them safely; the CDO on data access and model governance; the CRO and General Counsel on risk thresholds and escalation gates; the CHRO on how humans and agents share responsibility; and the CFO on spend and ROI. None of this is surprising. It’s just what each executive claimed in the insurance company meeting. What reframing the question accomplishes is that it shifts discussion from full ownership to creating granular and explicit decision rights. It is one thing for the CIO to say “I own the tech stack.” It is another to specify that the CIO owns foundation model selection, API integration standards that allow agent access to the company’s systems of record, a “kill switch” if agentic inference interferes with system latency, and so on. To make the shift from ownership to decision rights, companies will need to put into place a decision and accountability map. This entails identifying critical choices and assigning each to a specific executive. Tamping down jurisdictional frictions will require clear decision rights and unambiguous assignment of accountability. But a map is only as good as the process that maintains it, which means someone has to own the coordination layer itself. Where the Chief AI Officer Fits This returns us to the Chief AI Officer. Many companies are creating this role out of confusion about how to proceed, because they recognize either that AI ownership can’t belong to a single individual, or because no one on the current executive team has the expertise or bandwidth to lead the agentic AI adoption process. But again, a new role will succeed only if it has a defensible jurisdiction. A CAIO who “owns AI” will struggle, because AI touches everything. What a CAIO can own is the coordination layer. In practical terms, this means owning the decision rights map itself: creating it, maintaining it, identifying gaps, and driving for clarity when assignments to functional heads are unclear. In short, it means carving out appropriate jurisdictions for other functional leaders. And it means convening the relevant executives when a new use case doesn’t fit neatly into existing assignments, which I anticipate will happen frequently in the early years of agentic AI. This is a genuinely novel jurisdiction. No incumbent executive has ever owned the cross-functional orchestration of autonomous systems, but this is mission-critical as agentic AI becomes the new connective tissue in enterprise-scale organizations. The Inevitability of Instability None of these steps will eliminate jurisdictional competition, but they will channel it. If the conversation focuses on “who owns which decisions,” it will be healthy dialogue because it will move from abstraction to implementation-focused; it will more closely map to the needs of the organization; and it will lead to a framework that can be revisited as circumstances change. It is important to underscore that there is no permanent answer to the question of who should own which AI decisions. The technology is too broad, too significant, and too fast-moving for any settlement to hold. For the next few years at least, expect sustained jurisdictional competition, lots of shifts in decision rights, and equilibria that are dislodged almost as quickly as they form, as new waves of capability arrive. The insurance company CEO who convened that inconclusive meeting had a few options. He could wait to find clarity in a competitor’s approach or let the loudest voice win. Or he could acknowledge that every executive in the room was right, and that his job wasn’t to pick a winner but to spearhead a coordinated system in which each of them owns the decisions they are best positioned to make. That’s the real work of leading through jurisdictional upheaval of this scope and scale. It is to design the map that ties ownership to implementation, and then to revisit it every time the technology outpaces the current version of the org chart. Which, for the foreseeable future, will occur often.

Who in the C-Suite Should Own AI?

Summary.

Partner Center