Physics Fridays — Paper No. 32
- Robert Dvorak
- 9 hours ago
- 12 min read
Temperature Is Not a Decision
Why AI Works So Easily for Us—and Becomes So Much Harder Across an Enterprise
AI × IT × Human Intelligence (HI) × Operating Architecture = Economic Energy and Value
Executed through the assembly of Micro Operating Models for Macro Impact.
Author: Robert Dvorak
Founder, BlueHour Technology
August 14, 2026
Why This Matters to the C-Suite and Board
We already know AI can create meaningful value. Millions of people experience it in their own work every day. The larger opportunity is carrying that value across an enterprise without allowing complexity, cost and risk to grow faster than the value being created.
Integrating AI into a Personal Operating Model and integrating AI into an Enterprise Operating Model are fundamentally different undertakings. With one person, the human supplies much of the Operating Architecture. Across an enterprise, that architecture has to be deliberately designed.
There is a practical unit between the individual AI use case and enterprise-wide transformation. Micro Operating Models provide a bounded way to combine AI, IT, Human Intelligence and Operating Architecture around economically meaningful work.
As the rate of change increases, the ability to adapt becomes more valuable than the ability to predict. The objective is not to describe precisely what the enterprise should look like five years from now. It is to build an enterprise capable of continuously evolving as technology, economics, competition and risk change.
The economic objective should be larger than productivity. Revenue Growth, Cost Optimization, Risk Management and Operating Leverage are the outcomes that matter. AI productivity is valuable when it contributes to them.
I want to begin with the multiplication sign in the equation above. It is there for a reason.
The same is true of the title.
AI, IT, Human Intelligence and Operating Architecture do not create enterprise value independently. They operate as a system, and the performance of that system depends on all four.
Addition would be much more forgiving. If value were simply the sum of AI, IT, human capability and Operating Architecture, an enterprise could be excellent at three, weak at the fourth, and still expect most of the available return. That is reasonably close to how we have approached enterprise technology investment for decades. We add capabilities, integrate them into what already exists, train people, add governance and expect incremental value.
Multiplication changes the economics because every term affects the product. Extraordinary capability in one term cannot automatically compensate for weakness in another. A sufficiently weak term constrains the system, and a term at zero takes the product to zero.
Figure 1. Under an additive model, a weak fourth term costs about a third of the available return. Under a multiplicative one, it costs most of it.
AI gives us an unusually good opportunity to see this because the value of the technology itself is becoming increasingly difficult to dispute. Many of us experience that value every day.
The more interesting question is why something that can become so valuable to one person, so quickly, becomes considerably more difficult when we try to reproduce that value across an enterprise.
The Personal Operating Model
I use AI every day, and the value is obvious to me.
It increases the amount of work I can do and the range of work I can reasonably attempt. Research that once required hours can be completed much faster. I can test an argument, challenge an assumption, compare alternatives, analyze information and explore subjects at a depth and speed that would have been impractical only a few years ago.
Millions of people are having their own version of that experience.
We have integrated AI into our Personal Operating Models.
Consider what happens when one person uses AI effectively. I know the work I am trying to accomplish. I choose the AI tool. I provide context and determine what information is relevant. I evaluate the response against what I know. When something does not make sense, I challenge it. When I need another perspective, I ask another question. I determine whether the output is useful, whether I trust it and whether an action should be taken.
I remain accountable for the result.
There is an Operating Architecture in that interaction even though I have never formally designed one. I provide much of it myself. Decision authority is obvious because it is mine. I provide additional context when required. I recognize most exceptions because I understand the work. I can move rapidly between AI, other technology and my own judgment. Feedback is immediate, the number of interfaces is manageable, and if something looks wrong, I can stop.
The architecture is informal, but remarkably complete.
That helps explain why AI can be integrated into a Personal Operating Model quickly, inexpensively and with relatively little friction. For many people, the return becomes apparent almost immediately.
Now widen the operating environment.
Widening the Operating Environment
A large enterprise may have 10,000 people, 50,000 people or 200,000 people, each operating within some version of a Personal Operating Model.
Those people work across thousands of workflows, hundreds or thousands of applications, different data sets, organizational boundaries, management structures, regulatory requirements, customer commitments and decision authorities.
Integrating AI across that environment is very different from giving one capable person an AI tool.
The person asking a question may not own the underlying data. The person receiving an answer may not know where the information originated or how it was interpreted. An AI model may interact with systems owned by several different parts of the company.
Agents add another dimension because they can do more than provide an answer. They can act. An action can trigger another application, which can trigger another workflow or another agent. An output created in one part of the enterprise becomes an input somewhere else. A local decision can propagate across an operating environment that no individual person can see in its entirety.
Nothing about this means the AI became less capable when it entered the enterprise. What changed was everything surrounding it.
Integrating AI into a Personal Operating Model and integrating AI into an Enterprise Operating Model are not the same engineering undertaking at different sizes.
At the personal level, much of the architecture can remain implicit because the human supplies it.
At enterprise scale, the Operating Architecture has to become explicit.
Converting Capability Into Economic Work
Physics gives us a useful way to think about that conversion.
Energy is the capacity to do work. Having energy available does not mean useful work is being performed. A charged battery sitting on a shelf has energy. Water behind a dam has energy. A compressed spring has energy. In each case, the capacity to perform work exists, but that capacity still has to be converted into useful work.
AI creates extraordinary new capability. Enterprise IT represents enormous accumulated capability. Human Intelligence represents capability as well.
Economic value occurs when those capabilities change an outcome. Revenue grows. Cost declines. Risk is reduced. A customer stays. Capital moves to a higher-return use. A better decision gets made sooner.
At the personal level, the distance between capability and work can be remarkably short. I ask AI to help me research something, evaluate the research, make a decision and act.
The enterprise loop is much larger.
Between AI capability and enterprise economic value sit applications, infrastructure, data, people, workflows, controls, incentives, decision rights, policies, exceptions, dependencies and increasingly other AI systems.
This is why Operating Architecture belongs in the equation:
AI × IT × Human Intelligence × Operating Architecture
Operating Architecture governs how the other three interact. It determines how decisions move, where authority resides, when humans intervene, what happens when systems disagree, how exceptions are handled, what gets measured and how the enterprise responds when something moves outside its intended operating conditions.
When that architecture is weak, capability gets consumed by the enterprise equivalent of friction. Rework increases, coordination expands, decisions slow down, systems produce conflicting answers, people reconcile information manually and meetings multiply because the underlying operating environment cannot resolve the issue itself.
An enterprise can therefore add extraordinary AI capability without producing extraordinary economic value. It can also add AI capability while increasing cost, complexity and operational risk.
Local Value Is Not Enterprise Value
The most useful recent research has moved beyond asking whether enterprises are experimenting with AI.
BCG reported in July that nearly nine in ten CEOs are already seeing some cost or revenue benefit from AI in targeted areas. That matters because it confirms something many of us already experience personally: AI can create value.
The difficulty becomes more visible as the operating environment widens. Only 26 percent of the CEOs surveyed said their companies had embedded AI as part of a broader business transformation, while only 14 percent clearly defined the P&L impact for all AI initiatives. BCG also found that higher performers were roughly seven times more likely to redesign workflows and reshape the business end-to-end with AI.
Deloitte reported a similar pattern this summer. Forty-eight percent of respondents said their organizations had introduced AI without redesigning the workflows or roles surrounding it. Only 12 percent reported redesigning work at scale with a new Operating Model behind it.
Yesterday, Reuters reported that only 16 percent of Japanese companies in a Nikkei Research survey were using AI company-wide. Roughly 60 percent were using it only in limited areas.
Research released this week examining ChatGPT Enterprise activity across more than 1,500 organizations and 17 million messages also found wide variation in how rapidly and broadly companies use AI and concluded that organizations are still learning how to integrate it into their workflows.
I do not read this evidence as an indictment of AI. It points to a different conclusion: creating AI value locally and producing AI-driven economics across an enterprise are different accomplishments.
That matters because the next step is not necessarily another AI tool.
The Unit Between Personal and Enterprise
The Personal Operating Model gives us an important clue. AI works particularly well there because the operating environment is bounded enough for the human to maintain coherence.
The Enterprise Operating Model sits at the opposite extreme. It contains too many people, systems, dependencies and interactions to redesign coherently all at once.
We need an operating unit between them.
At BlueHour, we call it the Micro Operating Model.
A Micro Operating Model is a bounded part of the enterprise in which AI, IT, Human Intelligence and Operating Architecture are deliberately designed to work together around economically meaningful work.
It retains many of the characteristics that make the Personal Operating Model effective. The work has a boundary. The objective is known. The required information can be identified. Decision authority is explicit. Human judgment has a defined role. AI has a defined role. Exceptions can be observed. Outcomes can be measured.
But the Micro Operating Model is large enough to change the economics of real enterprise work.
The size of the unit matters. Make it too small and we return to optimizing individual tasks without materially changing the economics around them. Make it too large and the number of interactions, dependencies and exceptions begins overwhelming our ability to maintain coherence.
We are creating an operating environment within which AI can produce measurable economic value without complexity growing faster than the value being created.
As additional Micro Operating Models are assembled and made coherent, the Macro Operating Model itself begins to change.
Temperature Is Not a Decision
Now we can return to the title.
We can decide that we want a room to be 72 degrees.
That decision does not make the room 72 degrees.
Temperature is a macro property produced by conditions operating at the micro level. An individual molecule does not have the temperature of the room. The temperature we measure emerges from the behavior of an enormous population of particles.
A modern building gives us a practical version of the same physics.
The south side may be heated by the sun while an exterior door keeps opening on the first floor. A conference room may have twenty people in it while another room has been empty all day. Computers generate heat. Air moves through ducts. Heat moves through walls, windows, floors and ceilings.
The desired temperature can be perfectly clear while the local conditions continuously change.
Modern building systems use zones, sensors, feedback and local control. Each zone is small enough that its conditions can be understood and managed while remaining connected to the larger system. The building does not require every zone to behave identically. It requires the zones to operate coherently enough that the building produces the desired result efficiently.
The same logic applies to an enterprise.
A Board can establish an objective for Operating Leverage. A CEO can approve an AI strategy. A management team can design a Target Operating Model. Those decisions establish direction, but they do not produce the operating state.
That state is produced by what happens throughout the enterprise: millions of decisions, transactions, workflows, system interactions, exceptions and human judgments occurring continuously.
AI increases the speed, volume and potential consequence of those interactions.
The Macro Operating Model we want therefore has to be produced by the operating conditions underneath it.
Designing for Perpetual Evolution
This also changes how I think about long-range planning.
There was a comment this week that the five-year plan is dead.
I would go further. I am not convinced it was ever alive.
A five-year ambition can be valuable. Long-range capital planning is valuable. Scenario analysis is valuable. Knowing which markets we intend to compete in and what kind of company we want to become is valuable.
But no management team is particularly good at forecasting tomorrow, much less describing with precision the operating environment five years from now.
Think about what a five-year plan written in August 2021 would have needed to anticipate correctly about August 2026: generative AI, agents, inflation and interest rates, geopolitics, supply chains, energy requirements, data-center growth, labor markets and technologies that most Boards were not seriously discussing at the time.
The point is not that management failed to predict those things. Expecting that level of prediction was never particularly reasonable.
Enterprises are perpetually evolving because the environment around them is perpetually evolving. Customers change. Competitors change. Technology changes. Talent changes. Capital costs change. Regulation changes. Risk changes.
The Operating Model has to evolve with them.
As the rate of change increases, the ability to adapt becomes more valuable than the ability to predict.
That does not diminish the importance of planning. It changes what good planning should accomplish.
The Board should expect management to establish direction, allocate capital intelligently and define the economic outcomes the enterprise intends to produce. It should also expect the enterprise to recognize changing conditions and modify how it operates without requiring another massive transformation every time yesterday's assumptions become obsolete.
At BlueHour, we are not trying to design the Enterprise Operating Model of 2031. Nobody knows enough about 2031 to do that responsibly.
We are designing an Operating Architecture through which the enterprise can continuously evolve between now and 2031 while preserving economic discipline, operational coherence, Human Intelligence, resilience and truth.
Micro Operating Models make that practical because the entire enterprise does not have to be redesigned at once. Individual operating units can be measured, modernized, replaced and assembled as conditions change.
An enterprise that is perpetually evolving does not have a finished Operating Model, nor should it.
Coherence Becomes an Economic Variable
One hundred individually successful AI implementations do not necessarily produce a successful Enterprise Operating Model. They can also create one hundred new dependencies, interfaces and decision paths.
Every additional component has the potential to add capability, but every additional relationship also introduces complexity. As the system grows, the interactions between components eventually become as important as the components themselves.
This is why BlueHour has spent so much time on Constructive Interference, Fractality and Entropio. They address what happens as individually useful capabilities become increasingly connected and the relationships between them begin affecting the performance of the system itself.
We want Micro Operating Models to reinforce one another rather than interfere with one another. We need to understand where the system can flex as conditions change and where additional coupling begins consuming the value the new capability was intended to create.
At enterprise scale, complexity has an economic cost.
Operating Architecture should therefore do more than connect things. It should help the enterprise determine whether every additional increment of capability is producing more economic value than the complexity required to support it.
What BlueHour Does About It
BlueHour was built around this opportunity.
We work with enterprises to identify economically important work where AI can create measurable value and redesign that work as a Micro Operating Model rather than simply adding another AI capability to the existing environment.
We begin with the economics. What does the work cost today? Where is revenue being constrained? Where is risk accumulating? Where is complexity consuming margin? What should change? What economic result would justify changing the Operating Model?
Then we address what is required to produce that result: the AI, the existing IT and data, the Human Intelligence and decision authority that need to remain in the system, and the Operating Architecture required to make them work together.
We do not believe an enterprise needs another multi-year transformation program before it can act. Nor do we believe hundreds of disconnected AI pilots constitute an AI Operating Model.
Start with economically meaningful work. Establish the boundary. Build the Micro Operating Model. Measure it. Determine what worked, what did not and what changed around it. Then determine where the next Micro Operating Model belongs.
For some enterprises, the right place to begin is even more fundamental. AI, software, cloud, infrastructure and subscription spending may already be growing faster than the organization's ability to determine whether the portfolio is creating enough economic value to justify its cost.
That is why BlueHour developed MOM-001: Capital Discipline, a continuous BUY-HOLD-SELL discipline across applications, infrastructure, cloud, software, AI models, agents and subscriptions.
The objective is larger than cost reduction.
It is Operating Leverage: growing the economic output of the enterprise faster than the cost required to produce it while preserving the Human Intelligence, resilience, risk controls and truth required to operate safely.
If your enterprise is investing aggressively in AI but the C-Suite or Board cannot yet draw a clear line from those investments to Revenue Growth, Cost Optimization, Risk Management and Operating Leverage, give us one economically important piece of work.
We will start there.
What This Means on Monday
We already know AI can create value. Many of us experience that value every day through our own Personal Operating Models.
The opportunity is to widen that value without allowing complexity to widen faster.
That requires a practical unit of modernization, an Operating Architecture designed for continuous change, explicit Human Intelligence and decision authority, and economic measurement that reaches beyond AI activity to the performance of the business.
The enterprise does not need to predict precisely what it will look like five years from now. It needs to become very good at evolving.
We can establish the economic outcomes we want and define the direction of the enterprise. What we cannot do is predict every condition the enterprise will encounter along the way.
We have to assemble, measure and continuously evolve the Micro Operating Models capable of producing those outcomes as the conditions around them change.
We can decide that we want the room to be 72 degrees.
Temperature is not a decision.
Designed with Physics. Deployed with Economics. Determined by Humans.
Comments