Human Agency and Delegation / 02

Tolstoy, War and Peace, and the Illusion of Control in the AI Boom

Napoleon's orders and Kutuzov's attention to the battlefield offer a way into the uncertainty facing professionals in the AI boom: how can leaders recognize their limits without abandoning responsibility?

M.K.October 29, 20266 min read

Tolstoy reads in a bright loft, with a cultural magazine collage of city buildings and interconnected people

M.K. / FIELD NOTESHuman Agency / Technology History / 2026

When a manager says everything is under control, the words may express confidence. They may also mean the bad news has not reached the room.

Professionals in finance and technology know the pressure. Moving slowly risks missing an opportunity; moving quickly can bring poorly understood risks into a team. A particularly difficult admission is that the experience which earned us responsibility may no longer guarantee good judgment.

I want to approach that problem through Tolstoy's War and Peace. Recognizing that we cannot control the whole situation does not remove our responsibility for the choices we can make. That is my contemporary interpretation, rather than a historical statement by Tolstoy about AI or a management method extracted intact from his novel.

When the battlefield refuses the plan

In Book Ten, Chapter Thirty-Four, messengers repeatedly bring Napoleon requests for reinforcements at Borodino. Their reports do not fit his understanding of the battle. He eventually authorizes support, then changes which division should go. Orders travel, movements change, and time is lost.

Tolstoy depicts a troubling kind of failure. Familiar methods, experienced commanders and a record of success are still present, yet they cannot produce the expected outcome. For someone whose professional identity rests on being right, accepting this can be harder than revising a calculation.

The scene is literary characterization, not sufficient evidence for a verdict on the historical campaign or Napoleon's life. Its value is the opportunity to watch a person respond when reality stops confirming his expectations.

In my own interpretation, the useful question is whether we can name the conditions that helped a decision succeed. Can we also distinguish our mistakes from the limits imposed by circumstances when it fails? An explanation that assigns every gain to our insight and every loss to the environment is suspiciously comfortable.

Kutuzov deserves more than a leadership slogan

In the following chapter, Kutuzov attends to the expression and tone of the people reporting to him, as well as their words. In Tolstoy's portrayal, he understands that people facing death cannot be moved around like pieces on a board.

Yet this same character becomes angry and disputes reports. Selecting only his attentive moments would turn a complicated figure into a list of desirable managerial traits. We would be producing another hero to explain a situation the novel makes harder to simplify.

I would therefore resist assigning Napoleon and Kutuzov the roles of bad and good manager. More revealing questions concern whose information reaches a decision, which hesitations matter, and whose experience becomes visible only when the plan fails.

A workplace is also fundamentally different from a battlefield. Employees are not soldiers, and military analogy cannot justify sacrificing them. What we can borrow is an attention to people and information, rather than an entire logic of command.

An AI growth narrative is not yet a company plan

The IMF's October 2026 World Economic Outlook identifies both the potential gains from AI investment and productivity, and the risk of a reassessment of AI expectations. The 14 October press briefing also discusses the difference between expectations for near-term benefits and longer-term gains.

Those observations do not declare AI a failure or settle the prospects of any individual company. They remind us of a missing step: long-term value in a technology does not establish the case for every process, organization or adoption date.

Consider a hypothetical company using AI to help customer-service staff find information. This is an illustrative scenario, not a claim about a real company's results. Faster answers matter, but an increase in checking and correcting errors could consume the time saved. The useful questions concern correct outcomes and the total work involved, including work that happens after the model responds.

Calling for a small trial is not automatically responsible, either. Without a time limit or a clear account of what would count as success or failure, a trial can become an elegant way to defer a decision.

Put limits on the decision-maker, too

I would record why the project is being attempted, what results are expected, and what evidence would show the original judgment was wrong. This record should prevent the person proposing the plan from endlessly moving its goalposts. It should not become a tool for punishing staff who report an unwelcome result.

There are costs. Stating assumptions can make a proposal sound less certain. Listening takes time. Preserving a real ability to stop can make a decision less swift and impressive. If we are unwilling to acknowledge those costs, saying that we are monitoring progress may offer reassurance without any constraint.

In Part Two of the Epilogue, Chapter Nine, Tolstoy considers human action through relationships, time and causation when discussing freedom and necessity. We do not have to reduce that argument to a claim that individuals do not matter. It makes allocating all credit or all blame to one person less easy.

I have previously considered which judgments cannot be delegated along with work to AI. Reading the novel adds a more personal question: if the results show that I was wrong, who has the authority to make me change?

Common questions

Listen to the Mandarin conversation with M.K., Dorami and a literary reconstruction of Tolstoy for another way into War and Peace, decision-making and responsibility.

Do I need to have read the whole novel?

No. Napoleon's response to reports and Kutuzov's attention to the battlefield provide two entry points. They cannot represent the full range of the novel's concerns with families, war and ordinary life.

Did Tolstoy discuss artificial intelligence?

No. The connection to AI and workplace decisions is a contemporary interpretation. Any dialogue presenting a historical author in conversation about modern events should be identified as a reconstruction, rather than an authentic historical interview or quotation.

Does recognizing limited control provide an excuse?

It need not. Understanding constraints and explaining responsibility can proceed together: identify what could not be controlled, what choices were made, and when revision is necessary.

Is this an argument against every AI investment?

No. It is an argument for distinguishing a broad technological story from a specific decision. It does not recommend individual investments or predict returns.