Pilots are common and scaled agent use is rare. NKK's Oversight Map, Three-Meter Funding Model, Canvas and Maturity Ladder turn the operating-model research into owners, budgets and a scorecard.
Written for: CIOs, chief AI officers, COOs, CFOs, and transformation and HR leaders
12 chapters, 46 cited sources, 22 models and charts. The first 2 chapters are open; the rest unlock with the form.
of organisations are redesigning key processes around AI; 37% use it only at a surface level
In our reading, most large organisations now have access to the same models. What separates them is the organisation wrapped around those models: who decides, who owns, and who pays. This paper argues that an AI-native operating model is a short list of explicit assignments: a named owner for each value stream, class of decision, agent and unit of AI spend. The Center of Excellence (CoE) is the vehicle for making those assignments, and it should be designed to give most of them away.
The evidence supports a narrow claim. Adoption of AI reached 88% of organisations in the McKinsey data reproduced by the Stanford AI Index, yet scaled agent use is in single digits in nearly all functions [1]. Only 25% of Deloitte's respondents have moved 40% or more of their pilots into production [2]. MIT CISR's summary is that adoption is not value creation [3], and DORA finds AI amplifies the strengths and dysfunctions already present [4].
The evidence does not show which operating model wins: we found no controlled comparison, and most sources are self-reported surveys, several from firms that sell advice.
We offer five NKK tools built on that prior art: the Oversight Map, the Three-Meter Funding Model, the Governance Rhythm, the Operating Model Canvas, and a Maturity Ladder with an eight-dimension scorecard. The Canvas also consolidates two supporting devices, the Five Agent Records and the ownership stack. All are opinion and design, not validated results.
of organisations are redesigning key processes around AI; 37% use it only at a surface level
Access to capable models has stopped being a differentiator. Accenture's October 2026 argument is that when everyone has AI, the operating model sets organisations apart, and it defines an operating model as the way organisations turn strategy into performance [5]. The World Economic Forum, drawing on consultations with more than 450 executives, reaches the same place from another direction: scaling AI is "not a primarily technological challenge, but an organizational one" [6].
The numbers show the gap. Organisational AI adoption stood at 88% in 2025, and 79% of respondents said their organisation regularly uses generative AI in at least one function, in the McKinsey survey data the AI Index reproduces. (We use the 79% in Figure 4.3.1; the chapter highlights give 70%.) For agents, scaled use was in the single digits across nearly all business functions, and most respondents reported no agent use at all in most functions [1]. Other surveys paint a busier picture: KPMG's July to August 2026 pulse of 314 US leaders at firms with over $1 billion in revenue found 62% building, deploying or developing agents [7], and Deloitte found close to three-quarters of companies planning to deploy agentic AI within two years [2]. These figures measure different things (any pilot, scaled use in a function, a plan) and different populations. Our reading is that agents are widely piloted and rarely scaled, and we never add the percentages together.
What goes missing between pilot and scale is rarely the model. Deloitte found that only 30% of organisations are redesigning key processes around AI and 37% use it only at a surface level with little change to underlying processes [2]. MIT CISR puts the same point as a proposition: advantage "comes from one place only: the ability to take AI to production and scale it, with outcomes you can prove" [3]. BCG is blunter, saying becoming AI-first is "an organizational challenge, not a technical one" [8], though its 30/70 technology-to-people split is an assertion with no method on the page.
BCG's own measurement of the gap has moved. Its 2025 study of more than 1,250 executives found 5% "future-built" on a 41-dimension maturity score [9]; its September 2026 Applied AI Index, on a new sample of more than 1,300, reports 7.5% future-built and almost 50% now generating value from AI, and BCG does not compare the future-built share across the two samples, although it does compare some other metrics across them (agentic share of total AI value, 17% in the 2025 sample to 22% in the 2026 sample) [10]. We therefore use BCG's figures as dated, self-reported snapshots, never as a current success rate.
Every figure and regulatory statement in this piece traces to a numbered source below. Links open the original publisher in a new tab.
CoE research points to central platforms and rules with distributed delivery. What the evidence shows, what it does not, and four NKK models for placing each AI pattern.
A vendor-neutral reference architecture for enterprise agents: NKK's nine-layer stack, a pattern-selection matrix, a blast-radius gate, and the evidence on reliability, state, control and cost.
ISO 42001, NIST, the EU AI Act and MAS agree on the capabilities but not the operating model. NKK's tiering matrix, six pillars and evidence chain turn them into decisions, owners and proof.
