Hook
Imagine the change management worked. People across your organization are now using AI, capably and willingly. That is a genuine achievement, and it is still not the finish line.
What you have at that point is a lot of capable individuals. What you do not yet have is an organizational capability. The difference is structure: capable individuals each solve their own problems, in their own way, with their own tools, and the value stays trapped at the individual level. An organizational capability is when that individual skill is connected, governed and compounding, so the organization gets more than the sum of its people.
The structure that does this is the operating model: the ownership, the policy, the standard tooling and the sharing loop that turn scattered use into a managed capability. This lesson builds it. It is the difference between an organization where AI is a personal habit and one where AI is genuinely how the organization works.
The bottleneck is human, not technical.
Context
Ownership: someone is accountable
The first question an operating model answers is who owns AI. Not who uses it, who owns it: who is accountable for the tooling, the policy, the standards, watching the landscape and answering for how AI is managed.
When the answer is "everyone," the real answer is "no one," and an unowned capability drifts. Tools proliferate without review, the policy goes stale, nobody is watching the risks, and no one can answer a board's questions. Ownership does not require a large new function or a fashionable title. It requires a named person or role with the accountability and the authority to do the job. The single most common structural gap in organizational AI is simply that nobody owns it. Close that gap first; everything else in the operating model hangs off it.
Policy: the rules, on one page
The second part is the policy: the clear, written rules for how AI is used in the organization. The "AI Risk and Governance for Leaders" track covered what the policy must address, risk, data, accountability. The operating-model point is about its form.
A policy that is long, legalistic and unread protects no one. A policy that is one page, plainly written and genuinely known changes behavior. It should answer the questions a real employee actually has: what AI use is approved, what is forbidden, what data must never go into a general tool, what requires sign-off, and who to ask when unsure. Bias relentlessly toward brevity and clarity. The test of an AI policy is not its thoroughness. It is whether the people doing the work can recall what it says.
Standard tools: a sanctioned set
The third part is tooling standardization. Left alone, AI tools proliferate: every person and team adopts their own, and the organization ends up with dozens of overlapping tools, no consistency, a sprawling data-exposure surface and no leverage with any vendor.
An operating model defines a sanctioned set: the approved tools for the organization's main uses, on the proper account tiers, with the data terms the Governance track insisted on. Standardizing is not about restricting people for its own sake. It concentrates the organization's spend, its training, its support and its governance onto a known set, and it shrinks the risk surface. There is room for individuals to explore beyond the set, but the core of the organization's work should run on tools that are chosen, governed and known.
The sharing loop: making value compound
The fourth part is what most operating models miss, and it is what makes the capability grow rather than merely exist. When one person builds an excellent prompt, workflow or agent, does anyone else ever benefit from it?
In most organizations, no. The good work stays where it was made. A sharing loop is the deliberate mechanism that fixes this: a known place and rhythm where people contribute what they have built and discover what others have built. A shared library of prompts and workflows. A regular forum where people show what worked. It does not need to be elaborate; it needs to exist and be used. The sharing loop is what makes an AI capability compound, because each person's best work becomes a starting point for everyone else, instead of being rebuilt from zero again and again.
The inventory underneath it all
Holding the four parts together is a simple, current inventory: a living record of how AI is used across the organization, which tools, for what, by whom, touching what data. The "AI Regulation" lesson made the case for this from a compliance angle, and the operating model needs it for a management angle. You cannot govern, standardize, support or report on a capability you have not catalogued. The inventory is the unglamorous foundation the whole operating model stands on.
Steps
Step 1: Name the owner
Assign clear, named accountability for AI: a person or role responsible for the tooling, the policy, the standards, watching the landscape and answering for how AI is managed. This does not require a large function. It requires that the answer to "who owns this" is a name, not "everyone." Do this first.
Step 2: Write the one-page policy
Produce the AI policy as a single, plainly written page. Have it answer the real questions an employee has: what is approved, what is forbidden, what data must never be shared, what needs sign-off, who to ask. Test it by whether your people can recall what it says. If they cannot, it is too long.
Step 3: Define the sanctioned toolset
Decide the approved set of AI tools for the organization's main uses, on the correct account tiers with sound data terms. Standardize the core of the work onto that set to concentrate spend, training, support and governance and to shrink the risk surface. Leave defined room for exploration beyond it.
Step 4: Build the sharing loop
Create a deliberate mechanism for AI work to spread: a shared library of prompts, workflows and agents, and a regular forum where people show what worked. Keep it simple enough to actually be used. This loop is what makes the capability compound rather than stay trapped in individuals.
Step 5: Stand up the inventory
Create and maintain a simple, current inventory of AI use across the organization: tools, uses, owners, the data each touches. Keep it living, not a one-time document. It is the foundation the ownership, policy, standardization and sharing all depend on.
Recap
- Capable individuals are not an organizational capability. The operating model is the structure that connects, governs and compounds individual skill into a managed capability.
- Ownership comes first: a named person or role accountable for AI. When everyone owns it, no one does, and the capability drifts.
- The policy must be one page, plainly written and genuinely known. Its test is not thoroughness but whether the people doing the work can recall it.
- Standardize on a sanctioned toolset to concentrate spend, training, support and governance and to shrink the risk surface. Build a sharing loop so each person's best work compounds for everyone.
- Underneath it all sits a simple, living inventory of AI use. You cannot govern or manage a capability you have not catalogued.
You have the roadmap, the business case, the change management and the operating model. Now the operator's question: how do you actually deploy a specific agent or workflow into the enterprise, whether you built it or bought it, without it dying in pilot purgatory?