The difficult part of becoming AI-native is designing how experimentation becomes a capability the whole organization can use.
When a large organization decides to become “AI-native,” the first instinct is often to buy tools for everyone and hope useful behavior spreads.
That can create early momentum. It can also leave the organization with scattered experiments, uneven access to data, and no clear owner for turning an individual shortcut into a reliable capability.
KPMG is trying a more structural approach. On September 22, the firm announced its Client Technology & Innovation group, which will bring together the relationships and resources it uses to develop technology-enabled products. KPMG says the group will work at the “edge” of the firm, where new businesses can be tested and scaled before some of them are folded into the core.
The announcement interests me because it treats AI as an organizational design problem, not merely a software rollout.
A focused unit can have a clear mandate, appropriate resources, and accountability for a defined outcome. It can build deliberately rather than asking thousands of employees to discover the same opportunity independently. Within that kind of unit, task-specific agents may support different stages of the work without pretending that one universal assistant is the answer to every job.
Good organizations already divide human work this way. A person’s role, the sensitivity of the information, and the consequence of the decision should shape the tools and permissions they receive. People across the organization may benefit from AI, but they do not need identical context or autonomy.
This is why I do not believe every person needs every tool. I do believe every part of the organization will need to understand how AI changes the work around it.
KPMG says the new group will rotate emerging leaders through the edge businesses so they can carry AI and digital experience into future leadership roles. The firm also says CT&I will continue its broader work in adoption, upskilling, guardrails, and governance. Those statements describe KPMG’s intended model. They do not yet show whether the group will create successful businesses or improve the firm’s performance.
The central risk is separation. An edge unit can move quickly because it is protected from some of the core’s weight, but that advantage becomes a problem if the new products and operating lessons never travel back. The result would be an innovation island: impressive activity at the edge with little change in the institution.
The opposite failure is just as possible. If every experiment has to move through the full machinery of the core before it has been tested, the new group may inherit the very constraints it was created to escape.
The boundary therefore has to be designed with intention. The edge needs room to build and access to the expertise, relationships, and standards that make the larger firm valuable. The core needs a defined way to absorb what works without turning every pilot into a permanent exception.
A smaller organization does not need to copy KPMG’s structure. It may need one cross-functional team with a narrow mandate, or one department equipped with agents built for specific responsibilities. The principle is the same: separate exploration from production, name the owner, and decide in advance how a successful capability becomes part of normal operations.
The important question is not how quickly a company can put AI in front of everyone. It is which capabilities should be rebuilt around AI, who should own them, and how those capabilities will remain connected to the rest of the organization.
That is slower and more demanding than distributing licenses. It is also much closer to the work of building an AI-native organization.