AI and The Art of War

The Arrival of A New Model

The first day of the Iran conflict in February 2026 looked like the arrival of a new model for the art of war.

The model was clear and well-constructed: artificial intelligence had become essential to military and intelligence operations; the Pentagon needed access to frontier models; autonomous systems, AI-assisted targeting, and machine-speed intelligence were no longer side projects. They were becoming part of the operating system of modern warfare.

Then came the opening strikes. They were complex, coordinated, and precise. It was the kind of event that seemed to compress intelligence, surveillance, planning, logistics, and execution into one head-spinning set of sequences.

The new model had its first real-world demonstration.

A Model Is Not the Terrain

• The opening phase was a demonstration of AI-enabled compression. The promise was not science fiction, but the shrinking of time between seeing, understanding, deciding, and acting. The Pentagon’s Maven Smart System and related tools have been described as helping analysts synthesize surveillance, intelligence reports, targeting information, and battlefield planning at speeds human staffs could not match alone. (defensenews.com)

• The AI-warfare thesis appeared obvious. Better models, better sensors, better compute, better drones, better targeting, better coordination. A military keynote with explosions. A model that suddenly appeared to know everything.

• Then the terrain pushed back. Not all at once. Not as a single reversal. More like friction accumulating. Targets changed. The campaign became harder to read. The opening clarity gave way to the slower business of pressure, escalation management, shipping lanes, energy markets, diplomacy, and time.

• By mid-June, the center of gravity was no longer the opening strike. Axios reported that the U.S. and Iran had signed a memorandum intended to end the conflict, reopen the Strait of Hormuz, lift the U.S. blockade, and begin 60 days of nuclear negotiations. AP reported that shipowners had begun moving vessels through the Strait again after the agreement. (axios.com)

• The Strait of Hormuz was the unglamorous object in the story. It was not a model in the satisfying software-demo sense. It was geography, commerce, insurance, oil, ships, chokepoints, and leverage. The kind of thing a model can show clearly, but not fully control.

• That is the turn. The opening days showed what advanced intelligence and coordination can make possible. The later months showed that a system can still be shaped by older forces: terrain, incentives, bottlenecks, patience, and the places everyone depends on but no one entirely owns.

A Demo Is Not the Deployment

Something similar happens inside companies, though usually with fewer explosions and more dashboards.

An AI project often begins with a beautiful model. It shows the workflow, the bottlenecks, the handoffs, the approvals, the data sources, the obvious places where automation should help. Then the demo arrives and the model seems to come alive. The model summarizes what used to take hours. The agent drafts the report. The dashboard finds the anomaly. The workflow routes the approval before anyone has to ask.

For a moment, the company appears legible to itself and the future seems certain.

Then the model leaves the demo room and the project gets deployed. The data is a little less clean than expected. The process is a little less standard. The approval chain depends on people who were not in the pilot. The customer edge cases are stranger than the test set. The old spreadsheet is still, somehow, the real source of truth. None of this means the AI model failed. It means the deployment has entered the terrain.

That is where many AI projects become less dramatic but more revealing. The first question is whether the model can perform the task. The second is whether the task matters as much as everyone hoped. A company can generate more summaries, alerts, recommendations, drafts, and automated handoffs without necessarily changing the place where the business actually narrows.

Every company has some version of the Strait of Hormuz. It may be onboarding friction, sales-cycle length, customer trust, support volume, regulatory review, data quality, implementation capacity, gross margin, distribution, or executive alignment. It is the chokepoint everything still has to pass through. A good AI system may help find it, describe it, monitor it, or widen it. But a model, however impressive, is still not the terrain.

Orthogonal Take

This is not a sweeping lesson about the art of war or the future of technology. It is simply a vivid illustration of an old pattern.

A new system appears.
The first demonstration is astonishing.
The model becomes more detailed than anything people had before.
For a while, the model feels like control.

Then reality enters and keeps moving.

Good models matter. Better intelligence matters. Faster analysis matters. Better coordination matters. In business, a good AI system can reveal patterns people were too busy, too siloed, or too slow to see. AI can automate, accelerate, and scale processes in ways that were not previously possible.

AI and the systems built with it are genuinely new and extraordinary.

But the art and advantage still belong to the people who look outside the window frame and venture into the terrain.