People often borrow Sun Tzu for business when they want to sound ruthless. That has never been the part I find interesting. His account of strategy is intensely contextual. Before acting, a commander has to understand the ground and whether his position can be sustained. He also has to know what shape his own forces are in. A move that looks advantageous on a map can become disastrous when the route back is difficult or the army cannot support what the plan demands.
One of the categories in The Art of War is commonly translated as entangling ground. It describes a position that is easier to enter than to leave. I recognize that problem in the way businesses adopt technology. An offer can be genuinely useful at the beginning, then gradually weaken the customer's negotiating position as more of the organization forms around it. The initial purchase receives careful scrutiny because somebody has to approve a visible cost. Dependence arrives later, dispersed across ordinary decisions that rarely receive the same attention.
AI makes this especially easy. A capable model is available through an interface employees can use immediately, often at a price low enough to avoid serious resistance. People begin solving problems with it. As they work, they explain the business. They upload documents and correct the model. Over time they teach it which apparent rules have exceptions. Gradually, a general system becomes useful to that particular organization.
What the provider supplied was important, but it was incomplete. Employees added the history and operating context. Some of their contribution will remain in files the company can export. Other parts will be embedded in conversations or proprietary workflows. Some will survive only as habits nobody thought to document. When a better model appears, the organization may have to recreate much of that understanding before the new system becomes useful.
This is why I think lock-in is better understood as a change in leverage than as a narrow technical problem. Early in the relationship, the customer can compare alternatives with relatively little friction. Later, departure will interrupt established work and force the customer to reconstruct part of what it has learned. The contract may still allow an exit, but the organization has become less capable of exercising it. Implementation reports celebrate adoption and claimed time savings. They rarely ask whether a credible exit still exists.
There is nothing inherently foolish about relying on a provider. Organizations use specialist suppliers because building everything internally would be wasteful. What matters is whether the dependence is proportionate to the value and whether leadership understands what recovery would involve. The right to download a collection of files may satisfy a contract without giving another team enough context to use them.
The current enthusiasm for workforce replacement creates a related problem. AI will automate work, and some roles will change enough that fewer people are required. Leaders should examine those opportunities honestly. They should also look past the activities listed in a job description and understand the capability that has accumulated around the person doing them.
An experienced employee may know which of several procedures is current because she remembers the conversation that settled it. She knows why one customer received an exception and when that precedent should be ignored. If the normal workflow fails, she already knows where to look. This knowledge is easy to overlook because the organization receives it at the moment it becomes necessary. Competence often hides the complexity it absorbs.
A savings estimate can capture the work that disappears while missing the capability that leaves. The salary comes out of the model immediately. Costs created by weaker judgment emerge later as rework or a customer problem that nobody recognized early enough. Process maps contribute to the blind spot because they show formal activity while omitting the interpretation that keeps it functioning. An automated process may inherit the documented path and still require more supervision when circumstances depart from it.
The risk becomes more serious when provider dependence and workforce reduction reinforce each other. Employees spend months teaching an AI platform how the organization works, and management eventually treats the improved output as evidence that fewer of them are necessary. More context now resides with the provider while fewer people inside the company can interpret or reconstruct it. The files may remain available, but the understanding connecting them is divided between a departing workforce and a rented system. If the commercial relationship changes, the company has weakened its bargaining position and may no longer possess a credible route to recovery.
My concern for the people affected by these decisions belongs inside the business analysis. Empathy helps reveal contributions that accounting categories miss, but it also improves the prediction of how a change will behave. People notice whether their experience is being used or merely extracted before their role disappears. That perception affects what they share during implementation and whether they remain invested in making the transition succeed. A company that ignores this may lose knowledge before it even realizes the knowledge needed to be captured.
Sometimes the right decision will still be to reduce staffing or replace a provider. A serious analysis has to permit that conclusion. Preserving every current arrangement can restrict the organization just as surely as dependence on a new one, especially when technology has made the work safer or much easier. The important step is understanding what else moves when the visible cost is removed.
I have been applying this reasoning while building MOSAIK, a private AI operating platform I use as an independent business-transformation project. It began with a practical irritation. Each AI product encouraged me to rebuild knowledge and context inside its own environment. A better model could appear a month later, yet adopting it meant explaining the same history again. Although the documents were easy to move, I still had to recreate much of the work that made them useful.
My response was to keep the durable material in a knowledge layer under my control. Different models can use selected parts while the evidence and decision history remain in place. Human approval determines which conclusions acquire authority. This has made me less complacent about portability. Local storage creates dependencies of its own, and an open format may still be awkward without a particular application. Looking for those weaknesses gives me a more honest account of what I control and where I remain captive.
I use the term Intelligence Sovereignty for the broader objective. It concerns whether a person or organization can preserve accumulated intelligence and continue using it on its own terms. Owning the data helps, but files alone cannot preserve operational understanding. Access to several models also offers little protection when all the useful context remains trapped in one of them. The relevant measure is whether the organization can continue making competent decisions after a component changes.
Applied to an AI business case, this means examining what the proposed savings assume. Which capabilities are expected to survive the change, and where are employees supplying context the organization has never documented? I also want to understand how the provider's leverage will change as more knowledge enters its environment. If leadership describes the arrangement as flexible, the recovery plan should identify who could interpret an export and explain how the business would operate during transition.
The answers will differ. Strong documentation and internal technical depth may allow one company to change providers with little disruption. Another may discover that its supposed flexibility depends on a few people already planning to leave. A company may also decide that a provider's distinctive capability justifies considerable dependence, then prepare accordingly. The analysis should expose the position created by the decision without forcing every organization toward the same architecture.
Sun Tzu's attention to terrain is useful because it prevents us from evaluating a move by its immediate advantage alone. Business leaders navigating the AI transition have the same responsibility. Savings have to be considered in the context of the knowledge being transferred and the leverage that may follow it. They also depend on whether the people who understand the organization will still be able to respond when the planned route no longer fits the conditions.
The strongest decision may still require extensive automation within a proprietary platform, even if it supports a smaller workforce. Its quality will depend on whether leadership understands the position it creates. The organization must know which capabilities to retain, how its knowledge will remain intelligible, and how it can change direction without reconstructing itself. During rapid technological change, that continuing freedom of maneuver may prove more valuable than the savings used to justify giving it away.