Nobody should own AI

There is no shortage of executives putting their hand up to own AI. The CIO & CDO believe they do, it runs on data. The CTO believes they own it. It's technology! The CFO who controls the funding and the General Managers running the Pilots believe they own it too. While the board who owns the risk, positive and negative, is wondering why no one seems to own it.

The common conclusion is to assign an owner and end the war. Crown a new king, the Chief AI Officer, a role that moved out of obscurity to become a necessity within a single year. We've seen this before, and many are already predicting that those roles will disappear shortly when the novelty stops covering the costs.

Both the rush and the retreat miss the same point. The turf war isn't happening because ownership is contested. It's happening because the thing being fought over doesn't exist. "AI" is not owned by a line of business. It represents four distinct dimensions.

The four dimensions

Product AI: the intelligence in what you sell (products/services). This belongs with product and engineering leadership, and it always did. Nobody debates who owns the database in your product; the moment AI becomes a product capability; the debate should end the same way.

Work AI: the intelligence in how your organisation operates. This is the contested one, and the contest is revealing: it is a change problem wearing a technology costume. Its raw materials are workflows, roles, capability, and adoption. Not models. It is the largest of the four by value at stake and the dimension that is least like a technology program.

Enabling AI: the platform, data, models and tooling the other dimensions stand on. Shared infrastructure, owned like infrastructure, yet describing a semantic model of the whole organisation: by technology and data leadership, funded as a common asset, measured on what it enables rather than what it does.

AI Risk: governance, controls, the regulator conversation, and increasingly the question of what decisions and actions an autonomous system is allowed to do in your name. This belongs wherever enterprise risk already has teeth. Creating a parallel AI-risk structure outside your existing risk machinery builds a second, weaker immune system.

Four dimensions, four natural owners, four different clocks. The org-chart question dissolves the moment the decomposition is made. What remains is the genuinely hard part: the four dimensions are coupled. The product roadmap draws on the architecture; the work transformation creates the risk surface; and the coupling needs deliberate coordination. Coordination and collaboration by the executive team is what is required to bring the whole together.

Ownership follows the bottleneck

If the decomposition tells you what the dimensions are, one principle tells you where the contested one should sit: ownership follows the bottleneck, not the technology.

And on the evidence, the bottleneck in "Work AI" is not technical. Survey after survey comes to the same conclusion: executives overwhelmingly cite cultural and capability barriers, not model limitations. Tools are deployed; behaviour doesn't change; value doesn't move. The constraint is AI literacy, willingness, and a redesign of how teams actually work. Making it clear, we need to start with the people function and change impacts.

This logic recently produced one of the most interesting organisational move of the year. Atlassian expanded its Chief People Officer's role into Chief People and AI Enablement Officer. By placing internal AI transformation (including internal engineering and the company's data function) under the executive who owns culture and capability, they aligned ownership to “Work AI”. This reasoning aligns with the bottleneck argument: technology transformation and cultural transformation are inseparable, and when they are run in silos, tools get bolted onto workflows while nobody owns team productivity end to end. They also split their product technology leadership in two, one CTO for “AI in the product”, another for enterprise and trust. Sound familiar: it is the four dimensions, Product, Work, Risk - each with a named owner - and the architecture distributed deliberately between them. More on their ongoing AI transformation and lessons learnt.

Atlassian isn't the first, and probably not the last. Zapier made its people leader the company's first Chief People & AI Transformation Officer using similar logic. Adoption alone isn't transformation. True transformation is when AI reshapes how teams work together to create value (the company's operating system). Others like Coinbase and Block have taken a more radical approach to arrive at a similar destination. Rather than assigning “Work AI” to a function, they are embedding it into the structure itself, by flattened layers, player-coaches instead of pure managers, small pods directing agents. Ownership by design rather than by portfolio.

Adoption alone isn't transformation. True transformation is when AI reshapes how teams work together to create value (the company's operating system).

While both approaches are bold moves, it is still early with limited data to ring the bell. Atlassian's restructure arrived alongside deep job cuts (along with others aligning to a new AI-enabled organisational structure), and whether it proves transformation or theatre will take a year or more to know. The point is not that the people function is the answer everywhere. The point is that these organisations framed the question the right way. They found their bottleneck and put ownership on top of it, while most enterprises are still trying to crown someone over a kingdom that isn't one.

The throne and the pen

Where does that leave the Chief AI Officer? Mostly, as a symptom, likely to survive as long as the Chief Digital Officer (or until we've embedded AI into the way we do things). Many firms have taken one of two approaches. In one camp, data shows that chief executives have personally taken the AI decision (nearly three-quarters of them now), double the year before, with boards pushing faster than AI readiness can support. The other camp argues, more quietly, that centralised coordination matters far more than any specific title. Both are right, and neither supports the throne. The CEO holding the pen on AI's capital allocation is appropriate — it is a business-model question. But holding the pen is not doing the work, and a CAIO appointed instead of making the decomposition owns nothing but the ambiguity. A Chief AI officer with a handful of pilots changes the org chart without changing how the company runs.

There is a version of the role that works: the named custodian of the decomposition itself, the executive who keeps the four dimensions assigned, coordinated, and honest, especially early on, when there's been limited time to form good habits. That version is real, valuable, and probably transitional by design. The test of a good CAIO may be that the role eventually makes itself unnecessary; because ownership, once properly decomposed, has somewhere natural to live.

So, the next time the question circles the executive table — who actually owns AI here? Decline it. Ask the four questions it was hiding instead. Who owns the intelligence in our product? Who owns the change in our work? Who owns the ground they both stand on? Who answers when a system acts in our name? Those questions have answers. The original never did.