AI has arrived, accompanied by a tremendous wave of praise, anxiety and pessimism. For a system vendor, anxiety is inevitable. But after stepping back, one question becomes obvious: when has competition among us ever stopped?
One point needs to be separated from the noise. Most discussions about AI are too abstract. They talk about “disruption” and “paradigm shifts” without identifying which part of the cost structure has actually changed.
In my view, the first and most certain change brought by AI is the cost of building software functions. First drafts, boilerplate code, interface integration, documentation and test cases—work that once consumed a large share of engineering time—is being compressed. The consequence is not that one particular vendor will disappear. It is that functional convergence across the industry will come faster and more aggressively. A feature that differentiates a product today may be copied by a competitor in one quarter rather than one year.
That puts a direct question on the table: when functionality is no longer a dependable moat, what allows a system vendor to remain valuable?
- Stop Saying “Sell Management Concepts”—Sell a Position That Can Be Verified
A common industry phrase says: do not sell functions; sell a management philosophy. The direction is not entirely wrong, but it needs a reality check. If a philosophy cannot be verified, it is no different from an attractive line on a presentation slide. What customers pay for and use every day is still concrete functionality. I would rewrite the statement this way: every function must serve a coherent management position, and that position must be testable. I use three questions to judge whether a system truly has a position.
First, can it run an SOP without depending on one key individual?
If a process works correctly only because an experienced employee remembers everything, the SOP has not entered the system. It remains locked inside that person. A system should allow a new employee to follow the process without creating a major failure.
Second, does the system expose problems or conceal them?
Too many systems default to making the data “look normal”: aggregate it, smooth it and produce a presentable report. A system with a clear position does the opposite. It brings exceptions to the surface: how long has this step been blocked, is this employee overloaded, and is the gap between estimated and actual time continuing to widen?
Third, is data a natural by-product of doing the work?
If recording data is an additional burden, employees will bypass the system and return to chat tools and spreadsheets. The result is a shadow system and two versions of the truth. A good system allows data to accumulate naturally during normal operations instead of forcing people to enter it afterwards.
These are not abstract principles. They are criteria that can be checked directly against a product. If a system vendor cannot explain where it stands on these three questions, it is probably still selling a pile of functions without a coherent operating philosophy.
- AI Changes Different Things at Different Organizational Scales
For companies with fewer than 100 employees, coordination has traditionally depended on everyone being physically or organizationally visible to one another. AI can provide practical help by absorbing low-level repetitive work and increasing the effective output of a small team. Management, however, should invest in leading indicators rather than lagging ones: frequency of change, trends in estimation error and average blocking time can reveal where a problem is forming earlier than last month's financial statements. Small companies have little room for error, so seeing a problem early is more valuable than explaining it afterwards. For companies with 100 to 1,000 employees, the middle ground is more difficult and receives too little serious attention. The organization is already too large to rely on mutual visibility, yet its processes are not firm enough for everyone to trust the system unconditionally. Shadow systems and competing versions of the truth begin to appear.
AI cannot solve that underlying problem in this middle stage and may even accelerate the disorder, because it helps everyone create more content and more versions faster. The only durable answer is to make the system the single trusted source of truth. It is better to have a few views that are consistently accurate than ten views that are partly outdated. For enterprises with more than 1,000 employees, the challenge changes again.
In large enterprises, the enemy becomes organizational entropy. Information is repeatedly polished as it moves up the hierarchy, eventually forming a neat circle in which senior leaders make decisions against an improved version of reality. One important effect of AI at this scale is that part of middle management's role in routing information, consolidating status and translating between levels can indeed be compressed.
That does not mean middle management disappears. AI cannot take over motivation, talent development or exception handling. Organizations may become flatter, but two other requirements become more important. First, leadership intent must be expressed with exceptional clarity, because the more autonomy an execution layer has—whether it is a person or an AI Agent—the more clearly its objectives must be aligned. Second, validation and quality gates must become stricter, because incorrect outputs can also spread faster.
The role of a large-enterprise system must therefore change: less emphasis on merely recording who did what, and more emphasis on broadcasting intent clearly, validating outputs and maintaining a real-time context that both people and AI can read and update. “Real-time context” must be distinguished from the knowledge bases and data warehouses enterprises already have. A traditional knowledge base is written by people for people, updated after the fact and consulted only when someone chooses to look. The context described here is different in two ways: it updates as work happens rather than being archived afterwards, and one of its primary readers is AI. An AI Agent needs this context to understand the real state of work at the present moment before it can execute correctly. If a vendor talks about a “shared memory layer” without explaining these differences, it may simply be giving an old concept a fashionable new name. 3. An Uncomfortable Truth: Many Customers Still Buy Mainly on Functions and Price
After making the argument above, one uncomfortable fact must be acknowledged. In many real procurement decisions, customers primarily compare feature lists, price, the integration ecosystem and delivery speed. A system with a clear management position does not automatically win the order. It becomes a genuine advantage mainly under several conditions:
The customer has already paid a high price for shadow systems and unreliable data and understands the real cost of the problem;
The customer's industry processes are complex enough that a consistent operating model cannot be assembled by adding functions;
And the customer's management team is willing to make the corresponding process changes.
Conclusion
AI has not rewritten the underlying logic of business and management. Its impact is more direct and specific: it lowers the cost of building functions, which accelerates functional convergence and shortens the time during which “what the system can do” remains a differentiator. What can endure is a management position that can be explained and verified. It appears in whether the system reduces dependence on key individuals, exposes rather than conceals problems, and allows data to accumulate naturally through work. These qualities are difficult to copy in one quarter because they are not merely code; they reflect a product team's long-term judgment about how work should be organized. Management philosophy is cheap when no one is accountable for results. Only when a system vendor and an enterprise owner share responsibility for operating outcomes do the abstract claims fall away and the durable value become visible.