What Must a Soldier Still Know? Drawing the Competence Line in an AI-First Army

The U.S. Army is fielding AI-enabled sensing and targeting tools including the Maven Smart System and the Tactical Intelligence Targeting Access Node (TITAN), with XVIII Airborne Corps using Maven's Broad Area Surveillance-Targeting capability and Target Workbench and the 1st Multi-Domain Task Force adopting TITAN, an AI- and machine learning-enabled ground station that processes data from space, high-altitude, aerial, and terrestrial sensors. The service argues that adopting these systems requires drawing a "competence line" defining which tasks move to machines, which rely on human-machine teaming, and which competencies soldiers must retain without machine assistance. The capabilities have not yet been integrated across all combat formations, though the Army says they already offer operational advantages.

The modern battlefield is becoming more lethal and fast-moving, while the proliferation of sensors makes concealment increasingly difficult. Persistent sensing, precision-strike weapons, autonomous systems, and increasingly capable enemy artificial intelligence compress the time available to understand a problem, make critical decisions, and act on the battlefield. Soldiers and formations must therefore sense, decide, move, and strike faster than our peer and near-peer enemies. But competing at that speed requires more than new technology. It requires soldiers who can use increasingly capable systems without surrendering the knowledge and judgment required to fight when those systems are wrong, unavailable, or insufficient. Artificial intelligence is central to achieving that speed. We are pursuing AI-enabled sensing and decision tools to support everything from targeting to assessment to sustainment. These capabilities can help us act faster than our enemies, and we are already seeing that advantage in our formations. XVIII Airborne Corps, for example, has used https://www.lineofdeparture.army.mil/Journals/Field-Artillery/FA-2024-Issue-1/Airborne-Employment/ Maven Smart System’s Broad Area Surveillance-Targeting capability and Target Workbench to process imagery at scale, identify threat preparations, prioritize those threats for response, and move validated targets into fires workflows. The 1st Multi-Domain Task Force has received https://www.army.mil/article/282253/1st multi domain task force mdtf adopts titan a game changer in intelligence and targeting the Tactical Intelligence Targeting Access Node, or TITAN, an AI- and machine learning–enabled ground station that processes data from space, high-altitude, aerial, and terrestrial sensors to provide intelligence and targeting information. Although we have not yet integrated these capabilities across all of our combat formations, we have demonstrated their potential, and they already offer operational advantages. The challenge extends beyond acquiring better technology. AI will increasingly shape the relationship between targets and shooters, commanders and information, logisticians and supply networks, and staffs and the decisions they recommend. We need more than an AI adoption strategy. We need to draw a competence line and determine which tasks should move to machines, which should rely on human-machine teaming, and which competencies soldiers must retain without machine assistance. That line should shape how we train, educate, assess, and employ the force. Drawing the competence line requires leaders to make deliberate choices about competence. Commanders must identify what their soldiers must be able to do independently, where AI should accelerate or improve their performance, and where human judgment must be a safeguard against flawed AI-produced outputs. They must then train and assess against those distinctions. The question for leaders is no longer simply whether their formations are using AI, but whether their soldiers understand the tasks, assumptions, and decisions that AI increasingly performs for them. What AI Can Do for Soldiers The first generation of generative AI has dramatically reduced the effort required to process information and produce staff work. Large language models can search for information, summarize reports, draft orders, compare alternatives, generate code, and build briefings. Multimodal models can work across text, imagery, audio, video, and other forms of data. Retrieval-augmented generation systems can connect models to authoritative sources instead of relying only on information contained in their training data. These capabilities allow staffs to work at speeds that were impossible only a few years ago. Agentic AI goes further. An AI agent can https://www.lineofdeparture.army.mil/Journals/Field-Artillery/Field-Artillery-Archive/Field-Artillery-2025-E-Edition/AI-s-New-Frontier/ identify an objective, access relevant information, employ digital tools, complete a sequence of tasks, assess its progress, and adapt its approach with limited human direction. Multiple agents can divide the work—one can retrieve information, for instance, while another develops options for various force packages, and a third combines the results for consideration. Agentic AI can therefore perform significant portions of a staff process rather than simply assist with a single task. These capabilities are already changing staff work. They reduce the time a planner must spend searching doctrine or building initial planning products. An intelligence analyst may no longer need to manually correlate hundreds of reports before identifying useful relationships. A logistician may no longer need to manually assemble inventory, consumption, transportation, and threat data before recommending a resupply option. These capabilities are already being demonstrated at Army Materiel Command, which is using artificial intelligence to support sustainment decisions, including through Weapon System 360 https://www.army.mil/article/288960/the data advantage transforming army sustainment , which gives leaders visibility into parts inventories, suppliers, bottlenecks, and projected unit readiness across our supply chain. We must pursue these advantages. On a battlefield where decision time is compressed, we must embrace systems that help us make better decisions faster. We know, however, that speed does not always produce https://www.armyupress.army.mil/Journals/Military-Review/English-Edition-Archives/May-June-2026/Balancing-AI/ the best outcome; that requires judgment. That distinction matters because AI can make polished performance outpace the development of underlying expertise. A captain can enter a planning session with a polished course of action, synchronization matrix, risk assessment, and a commander update without possessing the expertise needed to evaluate the plan. Those products may not reflect an understanding of the enemy’s capabilities, the readiness of the formation, or the feasibility of the course of action. They may instead reflect the captain’s ability to produce a sophisticated output through an AI-enabled planning system. That is a problem we must address. What the Army Needs of Its Soldiers Effectively integrating and employing AI requires a comprehensive plan that spans doctrine, training, fielding, sustainment, and force design. Specialists will be necessary for model development, data architecture, cybersecurity, testing, and engineering, but AI-enabled work cannot remain the province of specialists if the force is expected to fight with it. A planner at the company or battalion level must be able to interrogate an AI-generated course of action. An intelligence officer must understand why an AI-enabled model assigns significance to a pattern and what evidence supports that assessment. A logistician must recognize when an optimization model relies on assumptions that no longer match battlefield conditions. Commanders must understand when automation increases tempo and when it creates vulnerability. That is the difference between possessing AI and being able to fight with it. Army history has repeatedly shown the danger https://asc.army.mil/web/news-been-there-done-that-behavioral-acquisition-part-two/ of allowing technological solutions to outpace the institutions required to employ them effectively. The Pentomic era demonstrated the risk https://history.army.mil/Publications/Publications-Catalog/Forging-the-Shield/ of designing formations around technological promise without the command structure, sustainment, protection, and institutional depth necessary to make them effective. In contrast, AirLand Battle provided a coherent concept for fighting that integrated changes in doctrine, organization, training, leadership, and technology. AI presents a similar institutional test. Drawing the competence line begins by determining where different kinds of work belong. Information retrieval, transcription, formatting, routine data comparison, first-pass synthesis, and other low-consequence cognitive tasks should increasingly move to machines https://www.armyupress.army.mil/journals/military-review/online-exclusive/2024-ole/ai-combat-multiplier/ when they can perform them reliably. We should not preserve manual work simply because it once served as evidence of expertise. Other military tasks will remain a partnership between soldiers and machines. Intelligence analysis is one example. AI can digest large amounts of reporting, resolve entities, correlate events, detect anomalies, and identify relationships faster than a human analyst. Maven already demonstrates part of that potential. Its algorithms can search large volumes of imagery and surface detections at a scale that would overwhelm a human team. Targeteers and geospatial intelligence analysts, however, still vet and validate those detections against the commander’s high-payoff target list, target selection standards, and attack guidance. The machine accelerates detection and correlation; soldiers determine whether its conclusions matter. The analyst must still identify the relationships that are operationally significant, recognize possible deception, assess threat intent, distinguish correlation from causation, and explain uncertainty to a commander. The analyst’s value increasingly lies https://www.lineofdeparture.army.mil/Journals/Military-Intelligence/Military-Intelligence-Archive/2025-July-December/The-Cognitive-Edge/ less in processing information and more in determining what matters and what conclusions the evidence supports. Operational planners will change in much the same way https://www.armyupress.army.mil/Journals/Military-Review/Online-Exclusive/2025-OLE/Modernizing-Military-Decision-Making/ . AI can search doctrine, compare resources, generate courses of action, model alternatives, identify inconsistencies, and test assumptions. The planner becomes more important for framing the problem, identifying what matters, rejecting unsound recommendations, and recognizing consequences the model may not capture. Artificial intelligence therefore does not simply reduce what soldiers need to know. It changes where expertise matters https://www.armyupress.army.mil/Journals/Military-Review/English-Edition-Archives/May-June-2026/Balancing-AI/ . Some skills will lose value because machines can reliably perform tasks that commanders once used as evidence of competence. Other skills will become more important because machines do not possess the judgment that thoughtful, experienced, and well-trained soldiers bring to military problems. Producing options may become easy. Determining whether an option is wise, feasible, or necessary will not. That judgment becomes particularly important when AI fails https://www.armyupress.army.mil/Journals/Military-Review/English-Edition-Archives/SO-24/SO-24-Artificial-Intelligence-Strategic-Innovation-and-Emerging-Risks/ . Large language models can hallucinate and present false information convincingly. Models can work from stale or incomplete data. Retrieval systems can access authoritative sources and still interpret them incorrectly. Training data and algorithms can introduce bias. Adversaries can attack AI-enabled systems through data poisoning, prompt injection, deception, and other methods. Soldiers can also become a source of failure if repeated exposure to machine-generated recommendations conditions them to accept AI outputs too readily. Keeping a human decision-maker in the loop is therefore necessary, but human presence alone is insufficient. Soldiers must possess enough expertise https://www.lineofdeparture.army.mil/Journals/Gray-Space/Archive/Summer-2025/Leadership/ to recognize when a machine is wrong, understand the consequences of acting on its recommendation, and intervene when necessary. AI may identify risk or recommend a target, but commanders remain responsible for accepting that risk and approving action. The technology will inform decisions; but it will not relieve a commander of responsibility for making them. The Way Ahead The competence line has an immediate consequence for how we develop and evaluate our people. Professional military education has traditionally relied on products as evidence of competence. Briefings, estimates, operation orders, white papers, examinations, and plans have served as useful proxies because producing them required soldiers to perform much of the underlying cognitive work. AI weakens that connection. A soldier can now produce excellent work without possessing all the knowledge historically required to create it. That does not make the work less valuable. Effective use of AI is itself a military competency. But a polished AI-generated product should not persuade us that the soldier who produced it understands the problem. Leaders therefore need to know both what their soldiers can do without AI and what they can accomplish with it. What soldiers can achieve with independent mastery—the ability to perform tasks without machine assistance—is important. What they can achieve with AI augmentation is also important.Neither is sufficient alone. For example, an officer with strong foundational knowledge of tactics and operations who cannot gain an advantage from AI may be too slow for the battlefield we expect to fight on. Further, an officer who produces exceptional AI-assisted work but cannot challenge its assumptions, identify a hallucination, explain a recommendation, or function when the system fails presents a different readiness problem. We should know which officer we have. Assessment should therefore expose reasoning rather than focus only on finished products. In our schoolhouses, a leader could receive an AI-generated course of action and be required to identify its weaknesses. An instructor could then change the disposition of a unit, remove a logistics node, or introduce contradictory intelligence and require the officer to reassess the recommendation. The officer should be able to identify assumptions within the model and explain what new evidence would change the decision. Only then should AI-enabled tools be restored and the officer be asked to improve the result. Our assessments should not be designed to prove that a soldier can outperform a machine. They should allow leaders to determine whether the judgment behind machine-assisted performance resides in the soldier. AI should also make our training more demanding, not easier. Generative models can rapidly create scenarios; AI agents can portray adaptive actors; and simulations can alter variables to generate new conditions. Instead of encountering only a handful of major decision problems during an exercise, a leader could work through many variations of a difficult problem. For leader development, much of the value may lie in repetition. The important measure is not how quickly a leader produces an operation order. What matters is how often the leader must make a meaningful decision. Leaders need repeated opportunities to frame problems, discover faulty assumptions, assess risk, confront conflicting information, challenge machine recommendations, experience failure, and adjust their approaches. AI can lower the cost of creating these repetitions without lowering the standard. We should use AI to increase opportunities for practice, not reduce the requirement to practice. The same principle applies to operational units. A combat training center should not be the first place soldiers encounter systems they are expected to employ or confront in combat. AI and autonomous systems need to become part of home-station training so soldiers can learn what works, what fails, how systems behave under degraded conditions, and when human intervention remains necessary. The Joint Multinational Readiness Center is training observer coach/trainers on AI-enabled capabilities so they can better prepare rotational formations to employ them. Combined Arms Command is likewise incorporating AI-enabled tools into command-and-control training and education, including field-grade professional military education. Soldiers should encounter these systems before their formations are evaluated under pressure at training centers. As we integrate these systems into training, education, and operations, the central question is not whether soldiers should use AI. They should. The harder question is what they must still know when AI can perform more of the work that once demonstrated competence. We must draw that line deliberately across the Army, and leaders must translate it into the training and standards of their formations. They must decide which competencies cannot be surrendered to the machine, build those competencies through training, and create conditions that require soldiers to exercise them both with and without AI-enabled systems. The soldier of the future does not need to outperform a machine at machine work. The soldier must understand war well enough to direct the machine, exploit its advantages, question its conclusions, and recognize when its solution is inadequate or wrong for the problem facing the formation. We are already determining how autonomy, AI-enabled sensing, and machine speed will change how we fight. We must be equally deliberate about determining what our soldiers must still know. Our responsibility as leaders is to ensure that as the machine becomes more capable, the soldier does not become less so. AI will move the competence line. We must decide where to draw it. Major Dorothy M. Reid currently serves as a special assistant to the vice chief of staff of the Army. A fixed-wing aviation officer, she has experience in aviation, intelligence, operational planning, and Army-level strategy. The views expressed are those of the author and do not reflect the official position of the United States Military Academy, Department of the Army, or Department of Defense. Image credit: Sgt. Sar Paw, US Army