Commander's Intent: Shifting Control from "Action Details" to "Purpose, Tasks, Boundaries"

In just a few short years, the way people use AI has undergone at least two paradigm shifts, which are particularly worth our deep reflection.

Initially, everyone studied “prompt engineering,” researching how to converse with AI: setting roles, providing examples, asking it to think step by step… Later, as AI became smarter, people realized that context information was more important than mere rhetoric, and the industry term upgraded to “context engineering,” emphasizing the provision of necessary data and constraints. Today, agents have become fully widespread, and the working method has evolved into “loop engineering,” also known as “harness engineering” or “agent orchestration.”

The core idea is that you can entrust an agent with a coarse-grained task, set the completion conditions, and let it run its loops autonomously—the agent itself breaks down tasks, finds information, performs operations, and conducts checks, refusing to stop until the standards are met [1].

For example, if you ask an AI to fix a webpage bug, there’s no need to specify the exact steps; a few simple sentences will suffice:

• I encountered this problem, your task is to fix it. You can read any code and run any tests.
• Completion is achieved when my specified criteria are met.
• However, you must not modify external interfaces, nor bypass security checks.

This is precisely how today’s AI-native engineers work. They don’t specify concrete steps, nor do they discuss any technical details; their communication style is like that of a leader. They tell the AI only three things: why, what constitutes completion, and what cannot be touched.

I dare say that AI will make us profoundly rethink human behavior patterns. If this is the most efficient way to assign tasks to an AI, and if even an AI doesn’t need to be micromanaged line by line, then why should a human adult employee be commanded and controlled in every minute detail?

If you manage your employees too meticulously, perhaps it indicates that your company’s business is overly simplistic. This article will explore a management philosophy consistent with the spirit of the AI era, namely “commander’s intent.”

Tracing Back: The Military Origin of Commander’s Intent #

This is a concept derived from the military. Its origins can be traced back to military reforms following Prussia’s disastrous defeat in 1806; by the mid to late 19th century, when Helmuth von Moltke was in charge of the General Staff, this philosophy gradually matured, forming “Auftragstaktik” (mission-type tactics). Moltke famously said, “No plan of operations extends with any certainty beyond the first contact with the main hostile force” [2].

Since plans are destined to fail, front-line personnel must be allowed flexibility. This philosophy was later adopted by the U.S. military and enshrined in Army doctrine, known as “mission command” [3] — the most crucial element of which is commander’s intent.

Don’t businesses often face the same problems as the military? In the early 21st century, British management consultant and historian Stephen Bungay, when researching why corporate strategies struggled to be implemented, summarized organizational failures as “three gaps” [4]:

• Knowledge Gap: The information you have cannot keep pace with rapidly changing reality.

• Alignment Gap: Subordinate actions deviate from superior intent.

• Effects Gap: Actual outcomes fall short of expectations.

In short, plans can’t keep up with changes. To bridge these gaps, decision-making authority must be delegated to those closest to the reality on the ground.

In fact, the Chinese also had this insight long ago. Sun Tzu’s Art of War states: “He whose general is able and not interfered with by the sovereign will be victorious” [5]. Ren Zhengfei also has a saying, “Let those who hear the gunfire make decisions” – this is no coincidence: in related speeches, he explicitly pointed out that Huawei “borrowed the organizational model of the U.S. Joint Chiefs of Staff” [6].

Mission command is by no means laissez-faire. A document from the U.S. Joint Staff specifically corrects this misconception: mission command is not less control—in a complex environment, enabling everyone to understand the commander’s intent is itself a more powerful form of control [3]. Low-level control is micromanaging actions; high-level control is unifying judgment standards.

Simply put, mission command means superiors define the intent centrally, and subordinates choose the means dispersedly.

Deconstructing “Commander’s Intent” #

Next, we delve into the components of “commander’s intent.”

Imagine you are a battalion commander, receiving the order: “Seize East Bridge tonight.” You lead your troops to East Bridge, only to find the bridge has been blown up by the enemy. What should you do? Should you just faithfully wait in place? Therefore, the order cannot be so simple.

Commander’s intent must contain three elements: purpose, key tasks, and conditions that define the end state.

Purpose: why this battle is being fought. For example, “The purpose of seizing the bridge is to open a passage for refugee evacuation and the main force to cross the river, and the main force must cross the river northward before dawn.” By clarifying this “why,” when you see the bridge destroyed, you will proactively look for a ford or build a pontoon bridge.

Key tasks: these refer to the essential core items that must be accomplished. For example, “Control the high ground on both banks, suppress enemy artillery observation points, and maintain the integrity of the bridge.” These are not action checklists; I don’t care how you achieve them, they are your objectives.

Conditions that define the end state: these are the clear and visible states that the situation must present after the mission is complete. For example, “Refugees have crossed the river, the bridge is passable, enemy fire cannot threaten the crossing point, and our follow-on forces are able to continue advancing northward.” This is a set of verifiable conditions, not a simple statement like “achieve victory.” The extent to which the battle is considered finished is clear and in black and white.

U.S. military doctrine also specifically requires: this intent must be short enough for the next two levels of commanders to memorize.

Hidden within the commander’s intent is another element called constraints, which defines what cannot be done. This includes legal and moral red lines, military discipline and rules of engagement, and hard constraints of the mission itself, such as “the bridge must be preserved, the refugee passage cannot be interrupted, the main force’s intent cannot be exposed”… These boundaries may be implicit in the key tasks and end state conditions, or they may not need to be explicitly written out. Constraints dictate which costs cannot be incurred to achieve victory.

In corporate operations, boundary awareness is equally indispensable. We can adapt this model to distill an enterprise version of commander’s intent:

Commander’s Intent = Why (Purpose) + What Success Looks Like (Tasks) + What Cannot Be Sacrificed (Constraints)

Purpose guides direction, tasks provide acceptance criteria, and constraints set guardrails.

Practice: How to Write an Effective Commander’s Intent? #

Let’s take a concrete example.

An internet company’s boss suddenly announced in the group chat: “Daily active users (DAU) must increase by 20% in the third quarter. All departments, submit your plans as soon as possible.” Such an order appears overly simplistic.

Does it mean any method to increase DAU is acceptable? Can we use check-in bonuses, buy traffic, or bombard users with pop-ups? Is sending SMS recalls to churned users considered harassment? If subordinates don’t understand the boss’s true intent and act on their own, they are “gambling” on the boss’s thoughts. If they guess right, it’s called strong execution; if they guess wrong, they might be accused of lacking a big-picture view, or even bear responsibility…

An excellent commander’s intent should be articulated as follows—

Why: To increase DAU, we must improve new users’ first impression, enabling them to genuinely accomplish something they couldn’t before.

What success looks like: New user retention rate for the second week, first task completion rate, and refund complaint rate must all reach target figures.

What cannot be sacrificed: Responsiveness and user privacy are non-negotiable; short-term number-boosting tricks like pop-up bombardment, incentivized check-ins, or buying zombie traffic are strictly prohibited. Beyond these, you are free to adjust guidance flows, copy, and default paths, and budgets within a certain limit do not require approval.

When you issue commands like this, all departments can clearly understand the direction of action. Especially when reality changes, such as a competitor suddenly announcing a free service, your employees can respond autonomously—because they understand what you genuinely want to achieve.

Implementation Challenges: Qualities an Organization Needs #

It’s not hard to imagine that for an organization to effectively adopt commander’s intent, it must possess certain inherent qualities. Mission command encompasses several core principles [3], from which we can distill five key conditions for executing commander’s intent:

1. Competence: Subordinates genuinely possess the ability to accomplish the mission.

2. Mutual Trust: Superiors trust subordinates to make sound judgments, and subordinates trust superiors to provide support.

3. Shared Understanding: Everyone shares a unified understanding of the current situation.

4. Risk Acceptance: Superiors acknowledge uncertainty and are willing to bear the cost of reasonable risks.

5. Disciplined Initiative: When necessary, subordinates must proactively adapt within the boundaries of the commander’s intent.

Comparing historical accounts from ancient and modern times, you’ll find how difficult it is to achieve these five points. Emperor Chongzhen and the officials and generals of the Ming Dynasty lacked basic mutual trust: if the superior blames the subordinate when things go wrong, how can the subordinate dare to undertake potentially winning but risky actions?

Let’s focus on the most crucial one: disciplined initiative. This is not what we commonly refer to as “exercising subjective initiative” but a very strict requirement. Doctrine explicitly states: when the original order is no longer suitable for reality, subordinates have a responsibility to adjust their actions within the commander’s intent and report to their superiors when possible—not “may” adjust actions, but “have a responsibility.”

An engineer receives a task to reduce the response time of a certain module to under 200 milliseconds. The original plan was to modify the caching mechanism. However, after some investigation, they discover that the real problem is not caching, but rather that upstream interfaces return a large amount of duplicate data during peak hours. If they merely followed orders, they would only make the cache very “pretty,” but the system would still be slow—whereas if they understood the intent, they should address the issue of upstream duplicate data or design a request merging mechanism, while also reporting upwards: the original plan is not suitable for reality; I recommend adjusting the direction.

So-called disciplined initiative means you can deviate from the original plan, but you must never deviate from the superior’s intent.

Of course, this also requires superiors to be magnanimous. Low-level organizations reward “not making mistakes,” cultivating a large number of “obedient” talents whose operating logic is, “I’ll do what you tell me; if it’s wrong, it’s not my responsibility.” High-level organizations, however, should reward those who make good judgments amidst uncertainty.

During the Austro-Prussian War in 1866, Prince Friedrich Karl of Prussia reprimanded a major. The major argued that he was merely following orders, and a military order was a royal command. The Prince responded with a line that has since entered military textbooks: “His Majesty made you a major because he believed you would know when not to obey his orders” [7].

“A general at the front may disregard orders from the sovereign.” True loyalty is not clinging to an outdated plan, but adhering to the purpose behind that plan even when the plan itself fails.

Broad Application: Diverse Scenarios of Intent-Based Governance #

Management based on commander’s intent means that in an uncertain environment, superiors define goals, boundaries, and trade-off principles, while subordinates autonomously choose means based on on-site information, and then maintain shared understanding through feedback. In fact, many fields are already doing this, and can do it even better.

For example, modern states manage monetary policy this way. Taking the Federal Reserve as an example, the statutory goals set by Congress for the central bank are essentially only two: maximum employment and stable prices [8]. As for when to raise interest rates or what tools to use, those are professional decisions for the central bank. The political system defines the intent, and professional institutions choose the means—the so-called central bank independence is not “experts can do whatever they want,” but “disciplined initiative.”

Most public governance should follow this principle: unified goals, defined red lines, rapid information feedback, and flexible on-site decision-making. Broadly speaking, China’s reform and opening-up is a super-large-scale intent-based governance: the central government provides the intent and boundaries, and how to implement it specifically is left for local areas to experiment first. “Crossing the river by feeling the stones” has a clear direction; everyone knows where the other side of the river is.

Another example is guiding graduate students in scientific research. A poor supervisor might say, “Go run this model and draw this graph,” and the student finishes without knowing why; a good supervisor will explain: which mechanism we need to test, what kind of results would overturn our hypothesis, which definitions cannot be changed… It is crucial to let students see “the big picture.” The core of scientific training is not to teach students to execute actions, but to enable them to learn how to identify problems, formulate hypotheses, and test theories.

Some parents micromanage their children in every detail, even down to what time to do homework, which subject to start with, where to put pencils, and how to pack a schoolbag. They think this is being responsible, but in reality, they are depriving their children of opportunities to practice judgment.

This is not to say you should be laissez-faire, but you can certainly state your intent: you can go out and play, but be safe, be on time, and contact me immediately if there’s danger; arrange your homework yourself, but it must be completed before bedtime tonight, legible, and with corrected mistakes. If you completely map out the route, the child learns obedience; if you clearly explain the intent, the child can develop disciplined initiative.

The Value of Intent: More Profound Than Metrics and Loyalty #

According to the standards of mission command, most bosses in the world probably don’t know how to assign work. The reason they can continue to be bosses is, firstly, because employees are “sensible”—they observe subtle cues, mentally fill in the boss’s unstated intent, and automatically complete tasks satisfactorily; secondly, it’s by deflecting blame downwards. But AI won’t humor you: if you don’t specify the “why,” it will invent one; if you don’t provide trade-off principles, it will choose arbitrarily; if your acceptance criteria are vague, it will perfunctorily deliver. The same model that thrives in someone else’s hands might flounder in yours, and that’s when you’ll truly realize whether you understand how to lead.

Expressing intent is becoming an increasingly valuable, possibly the most important, skill.

Only humans have intent. AI has no dissatisfactions with the world; whatever intent you give them, they will execute. It is precisely because we believe the current reality is not good enough that we need to translate a vague desire into an actionable intent.

I believe one of humanity’s most promising aspects is that “intent can be understood.” This world has readability problems, Goodhart’s Law, unintended consequences, and the banality of evil—but if you can bypass bureaucracy and sit down face-to-face with someone to talk, they can actually understand your intent.

Once the intent is understood, they can transcend all metrics.

Metrics, processes, and systems are all human-made means, all editable, whereas intent is something more fundamental.

Farewell to Micromanagement: Loyalty to Ideals, Not Orders #

Finally, let’s talk about a historical event. The Song Dynasty, learning from the chaos of previous dynasties, feared nothing more than generals acting independently outside the capital. Emperor Taizong Zhao Guangyi invented a method called “formation diagrams” (阵图). During campaigns against foreign enemies, the imperial court would repeatedly draw up these diagrams and send them by fast horse to the front lines, and generals had to deploy their forces according to the diagrams. But battlefield situations change rapidly; how could one micromanage from afar?

However, even though the generals knew the diagrams did not fit the actual battlefield conditions, they still fought according to them—if they lost following the diagrams, it was fate; if they won by not following the diagrams, it was an act of defiance against imperial decree!

Later, the censor Tian Xi, unable to bear it any longer, submitted a memorial pointing out: since generals had been appointed, they should be “entrusted with responsibility,” and there was no need to bestow formation diagrams or strategic plans [9]. But the emperor did not listen. As a result, the Northern Song Dynasty suffered repeated defeats against the Khitans.

Today, no emperor bestows formation diagrams, but have you seen a 17th version of a requirements document, a Gantt chart precise down to the hour, or sales SOPs with even the exact wording prescribed?

Behind micromanagement lies loyalty and precaution, a pre-modern management model.

Loyalty to commands often serves to absolve oneself of responsibility; loyalty to intent means taking responsibility for one’s judgment. And intent can be traced further upwards: what larger intent does this intent serve? And what ideal does that larger intent serve? Traced to its end, loyalty is no longer loyalty to a person, but loyalty to the ideal that deserves to be upheld.

If one day you discover that the superior giving you orders has become corrupt, would you still be loyal to them? The true loyalty of modern people is not to commands, nor to individuals, but to ideals.

Notes

[1] Anthropic. “Effective Harnesses for Long-Running Agents.” Anthropic Engineering Blog, November 26, 2025. https://www.anthropic.com/engineering/effective-harnesses-for-long-running-agents

[2] Moltke, Helmuth von. “Über Strategie” (1871), in Moltkes Militärische Werke.

[3] Department of the Army. ADP 6-0: Mission Command: Command and Control of Army Forces. Washington, DC, 2019. https://armypubs.army.mil/epubs/DR_pubs/DR_a/ARN34403-ADP_6-0-000-WEB-3.pdf；Joint Staff J7. Insights and Best Practices Focus Paper: Mission Command. 2nd ed., 2020. https://www.jcs.mil/Portals/36/Documents/Doctrine/fp/missioncommand_fp_2nd_ed.pdf

[4] Bungay, Stephen. The Art of Action: How Leaders Close the Gaps between Plans, Actions and Results. London: Nicholas Brealey, 2011.

[5] The Art of War, “Strategic Attack” chapter: “Thus, we may know that there are five essentials for victory: He who knows when he can fight and when he cannot will be victorious. He who understands how to use both large and small forces will be victorious. He whose officers and men are animated by the same spirit throughout all his ranks will be victorious. He who, prepared himself, waits to take the enemy unprepared will be victorious. He whose general is able and not interfered with by the sovereign will be victorious.”

[6] Ren Zhengfei, “Let those who hear the gunfire make decisions,” speech at Huawei Sales and Service System Struggle Awards Ceremony, January 2009; republished by Sina Tech, March 18, 2009. http://tech.sina.com.cn/t/2009-03-18/10512920001.shtml；Tencent News, “Let the front line directly call for artillery fire,” January 20, 2024. https://news.qq.com/rain/a/20240120A08FBJ00

[7] Anecdote of Prince Friedrich Karl (Austro-Prussian War, 1866): “His Majesty made you a major because he believed you would know when not to obey his orders.” See Muth, Jörg. Command Culture; War on the Rocks, “When Not to Obey Orders,” July 2019. https://warontherocks.com/2019/07/when-not-to-obey-orders/

[8] Board of Governors of the Federal Reserve System. “What economic goals does the Federal Reserve seek to achieve through its monetary policy?” https://www.federalreserve.gov/faqs/what-economic-goals-does-federal-reserve-seek-to-achieve-through-monetary-policy.htm

[9] Song Taizong’s “Commanding Generals from Afar” and the “Complete Battle Formation Diagrams”; Tian Xi’s memorial: “Since generals have been appointed, please entrust them with responsibility, and there is no need to issue formation diagrams or strategic plans.” See The Paper · Private History, “The ‘Complete Battle Formation Diagrams’: Song Taizong’s ‘Armchair Strategy’.” https://www.thepaper.cn/newsDetail_forward_24083727