Reading NTT DATA’s article The real AI opportunity: Reinvent how your organization operates felt familiar - not because of its transformation language, but because of the structure beneath it. The question is no longer where to insert AI, but how the operation should work when AI is one capability among others. That closely matches my view: AI is not a layer over an old process, but a component inside a designed system of thinking and action.
“Don’t automate the old process - reimagine it” captures the point. Organizations still add a model to a task and call it transformation while assumptions, information flows and decision rights remain untouched. At best, the old machine runs faster. At worst, its confusion is automated at scale.
NTT DATA proposes an end-to-end view: what would the process look like if designed today? Its other important observation is that “Not everything needs AI.” Deterministic work may be better handled by ordinary software; AI matters where ambiguity and interpretation enter. The model is not the starting point. The work comes first, and the value comes before the work.
This is where the resemblance becomes substantial. I think it is genuinely important to avoid making the maximization of AI a goal in itself. The real objective is always to build a system capable of creating something good, valuable and sustainable, and to do so reliably. AI is a commodity, not a strategy and not even a tactic. It is more like electricity: something the system may need, but whose value comes from what we accomplish with it, not from the mere fact that we use it. What must be understood and verified? Which rules should remain deterministic? What are we trying to accomplish, and why? When does human judgment add value, and who - or what - may turn a finding into action?
From process redesign to cognitive-system engineering
The term that best exposes this thinking is cognitive-system engineering. Process engineering describes activities; system engineering connects components and constraints. Cognitive-system engineering designs how a system creates context, tests interpretations, converts understanding into action and learns. It asks not only what happens next, but why that step can be trusted.
This is where my position goes beyond the article. Its main unit is the operating model. Mine begins with purpose and the cognitive model needed to fulfil it: understanding, evidence, reasoning tasks, division of capabilities, validation, control and authority. The workflow expresses this deeper logic. If it is weak, rearranging boxes and deploying agents will not repair it.
To illustrate cognitive-system engineering, I created the Sales Motion as a practical example of a fully workable, evidence-gated AI system. It contains the necessary steps, an explicit cognitive model for each key phase, and supporting capabilities, like 'Simulator' at step 3 (see the picture below), that operate both horizontally across the motion and vertically within individual phases. This system is not a prompt chain manufacturing sales material – it is an architecture built around explicit cognitive models for turning market understanding into commercial movement. Foresight builds a view of change. The Company Summary creates context. The Pitch establishes why a conversation should matter. Buying Insight and Hypothesis turn relevance into a testable interpretation of the customer’s situation. Action and feedback carry it into reality:
Each module has a cognitive responsibility, yet its value depends on the relationships. Foresight without company context becomes commentary. A Company Summary without a point of view becomes a fact sheet. A Pitch without a grounded hypothesis becomes persuasion detached from reality. An insight that cannot be challenged becomes a confident assumption. The architecture preserves the thread from external change to a defensible next move.
AI can scan information, connect patterns, propose explanations and challenge inconsistencies. Deterministic mechanisms enforce structures, boundaries and thresholds. Humans frame purpose, judge significance, test hypotheses in real relationships and change direction when reality contradicts the model. Reliability emerges from the composition - not from one participant doing everything.
I share NTT DATA’s insistence that governance cannot be added afterwards, but I would state the principle more strongly:
Security and privacy gates, evidence authorization, provenance requirements, deterministic thresholds, packet-quality gates and publication authority are not governance documents or decision forums sitting beside the system. They are inside the system. They constrain what the system can access, accept as evidence, infer, promote, publish and act upon. This turns governance from advice into enforceable boundaries. Generated narrative can support judgment but cannot grant itself authority. No finding may cross a missing source, failed quality gate or absent approval because its language sounds convincing. The architecture itself says no.
The leadership difference is not a minor detail
The most important difference concerns leadership. NTT DATA presents CEO sponsorship and C-suite alignment as central conditions of transformation. This is understandable: large changes require resources, aligned priorities and the removal of organizational obstacles. Leadership can accelerate change - or suffocate it. I do not dispute the practical value of sponsorship.
What I reject is the slide from that practical observation into a philosophy: that AI demands a new leadership model, transformation begins at the top or executive conviction is the cognitive origin of change. Technology creates no such law. AI does not make hierarchy wise, centralization intelligent or intent equivalent to understanding. A CEO can authorize investment without understanding the work. A specialist may see the systemic possibility first. Good ideas do not become valid because they descend from above, and weak architectures do not become coherent through C-suite support.
For me, leadership is a function inside the system, not a mythology above it. Setting direction, creating conditions, resolving conflicts, allocating resources and accepting accountability are important responsibilities. They do not entitle leadership to monopolize sense-making. The question remains: what capability, information and authority are needed, and where should they reside?
The distinction matters because language shapes architecture. If transformation starts with one leadership decision, we design a cascade: leadership decides, management translates, employees adopt and technology executes. In a cognitive system, signals can originate anywhere, interpretation is distributed and evidence moves upward as well as sideways. Decisions sit where responsibility and relevant understanding meet. Leadership remains necessary, but it is not the organization’s entire intelligence.
The Sales Motion embodies this philosophy. It does not need a heroic leader to manufacture certainty and push it through a funnel. It needs different forms of intelligence to cooperate: market sensing, customer understanding, strategic interpretation, creative framing, critical testing and human interaction. Authority must enable action, but no status should protect a hypothesis from evidence. People must be able to question assumptions, and feedback must change the next step.
The common ground—and the step beyond it
The similarities with NTT DATA are real. We agree that isolated AI use cases leave value on the table: broken processes should not simply be automated, deterministic technology often beats AI for deterministic work, and value appears at the end-to-end system level. This is genuine convergence toward system-centered thinking, not shared fashionable vocabulary.
The difference is where the journey leads. The article leans toward greater agentic autonomy and top-level sponsorship. My destination is not maximum autonomy or leadership-led AI as an ideology. It is greater cognitive capability with bounded authority: humans, deterministic software, and probabilistic AI doing what each does well, with outputs connected but not confused, claims challengeable, consequences never outrunning evidence. And this is the promise of cognitive-system engineering. It replaces “How much AI can we deploy?” with “What kind of system can understand and act on this purpose reliably?” The answer may be a powerful model, a simple rule, visible uncertainty, a conversation or a human decision. The intelligence lies in the arrangement.
NTT DATA’s thinking resembles mine. The discussion should move beyond tools and technologies to the architecture of work, value and purpose. I would take one more step: we are not merely redesigning how we create value around AI. We are learning to engineer how organizations think and to understand the cognition of the people within them - without surrendering judgment to a model or imagination to a hierarchy.
Cognitive-system engineering is an exploration of how humans and organizations think. And I would rather be an explorer - an adventurer - than an archaeologist of existing processes. What about You?
Reference: Craig Vaughan, “The real AI opportunity: Reinvent how your organization operates,” NTT DATA, 9 September 2026. The short phrases quoted in this essay are attributed to that article; the interpretation and comparison are the author’s own.