Conversational AI as a Cognitive Accelerator

Article Summary:

AI is already embedded in military systems, but conversational AI built on large language models (LLMs) represents a qualitatively new step. By allowing operators to query vast operational databases in natural language and rapidly generate courses of action (COAs), these tools accelerate the OODA loop (Observe–Orient–Decide–Act). The article traces this through the U.S. Maven Smart System / Palantir AIP ecosystem, showing how it has evolved from image analysis (2017) to full decision-support platform (2024–26), now deployed across all major U.S. combatant commands and NATO allied operations. This directly serves the subsidiarity principle at the heart of mission command — if units can understand the situation and generate options faster, they can act more effectively within the commander’s intent.

In an operational environment saturated with data, the question is no longer simply how to make decisions faster than the adversary, but how to retain control over decision-making in a system where information, its processing, and its flow are increasingly automated.

Artificial intelligence is now emerging as a major driver of this transformation. Already widely integrated into military systems, it enables the exploitation of unprecedented volumes of data—accelerates analysis, and proposes courses of action in near real time.

A more subtle evolution is taking shape, however: the emergence of conversational artificial intelligence capable of interacting directly with operators in natural language and serving as an interface between increasingly complex technical architectures and human decision-making.

This dynamic is not merely a technological breakthrough. It strikes at the very heart of contemporary command structures. By facilitating access to information, accelerating the generation of options, and reducing the time required to understand the situation, these tools reinforce the principles of Command by Intention (CPI).

But they raise a central question: to what extent does this algorithmic mediation transform the way decisions are made, and thus, ultimately, the way command is exercised?

This three part article explores this tension, focusing specifically on the role that conversational artificial intelligence could play in operational preparation and force training. For it is precisely in this space, between learning, experimentation, and testing of tools, that the balance between accelerated decision-making, technological dependence, and the preservation of human judgment is taking shape today.

This article (published in three parts) thus assesses the potential that such a development holds in terms of operational readiness and the transformation of command, but also of the implications regarding dependence and the safeguards it entails—implications of which the key stakeholders are well aware.

Toward an Acceleration of the OODA Loop: The Emergence of a Military “Cognitive Assistant”

From military artificial intelligence to the conversational interface

While artificial intelligence is already widely used within the armed forces, it encompasses several categories of complementary technologies.

For analytical purposes, five main types of AI can be distinguished:

• Analytical AI, which enables the processing of vast volumes of data, the detection of patterns or activity trends within them, and the generation of classifications—or even predictions—based notably on satellite imagery or sensor feeds.
• Generative AI, which is capable of producing new content (text, code, images, or syntheses) based on the data on which it has been trained.
• Conversational AI, which is a specific application of generative AI and allows users to interact in natural language with complex systems to query operational databases, quickly generate analyses, and explore courses of action.
• Decision-making AI, which aims to optimize choices or plan resource allocation within a given set of constraints.
• Finally, autonomous AI, which is capable of acting directly in the physical environment, for example within drones or robotic systems.

NATO’s strategic documents indicate that artificial intelligence must support a wide range of military functions, ranging from improving situational awareness and the exploitation of big data to decision support and the development of autonomous systems. In practice, these applications correspond to different stages of information processing: threat perception and detection, data analysis, generation of course-of-action options, and, in some cases, automated action. In the most recent architectures, these functions tend to be combined within integrated platforms capable of processing vast volumes of information and providing decision-makers with recommendations in near real time.

It is in this context that the integration of conversational artificial intelligence tools based on large language models (LLMs), already underway within the U.S. Department of War (DoW “Department of War”), notably through the combination of the Maven Smart System and the Palantir Artificial Intelligence Platform (AIP), and now being explored within NATO, could mark a new milestone in the operational readiness and training of armed forces. By enabling military personnel to directly query vast datasets using natural language and rapidly generate multiple tactical choices or operational options (COA for “Course of Action”), these tools aim in particular to accelerate the decision-making cycle on a battlefield increasingly saturated with information.

This new technological advancement could thus serve as a major lever in amplifying the principle of subsidiarity, which lies at the heart of contemporary command structures and holds particular significance in the air domain, where the speed of engagements, crew autonomy, and the need to make decisions in uncertain environments make it a de facto practice.

Whether it be “mission command” within NATO and the U.S. armed force or “Command by Intention” (CPI) taditionally associated with the French Army, these command models or philosophies rely on the ability of engaged units to quickly assess the tactical situation and autonomously determine the course of action necessary to achieve the desired effect.

They are embodied in concepts such as “distributed operations,” as conducted by the U.S. Marines, or NATO’s aerial dispersion doctrine known as “ACE” for “Agile Combat Employment ” (in France, the acronym “French ACE” has replaced “MORANE” for Reactive Deployment of Air Power. General Jérôme Bellanger, Chief of Staff of the Air and Space Force, describes the three pillars underpinning C2 within the third dimension as follows:

Our C2 is organized according to a threefold principle that is quite specific to aerospace military power. The first is common to all armed forces: it is centralized command, in other words, the unity of command.

The second, which is more specific to us, is the principle of distributed command. The Air and Space Force (AAE) is a global force. This is a reality that is actually quite recent, and one we owe to the modernization efforts of our fleets undertaken over the past decade, exemplified by the trio of the Rafale fighter jet, the A400M transport aircraft, and the A330 MRTT tanker. Today, we are capable of shifting our operational focus from one theater to another, from one point to another across the globe, in less than 48 hours. This reality necessitates the principle of distributed command and the non-affiliation of our fleets with any particular regional command. This is the prerequisite for our responsiveness.

Finally, the third and final principle of our C2 is decentralized execution. It fosters responsiveness and initiative, which are the fundamental qualities of aerospace military power: “Flexibility is key to Airpower.” This phrase, repeated at every briefing by all allied aircrew, aptly describes this culture of adaptability that forms the ethos of the AAE. “For us, initiative is the ultimate form of discipline.

Their gradual extension to all environments, within a multi-environment, multi-domain (M2MC) or multi-domain (MDO) framework, now reinforces their scope.

As highlighted in an analysis published by West Point’s “Modern War Institute” as early as 2022:

The increasing complexity of decision-making and mission execution are among the inevitable implications of MDO. Effectively coordinating different actions across multiple domains, including cyberspace and the electromagnetic spectrum, is akin to playing chess on several boards stacked one on top of the other, where every move made influences the possible moves on all the others. Here, complexity is not additive; it is multiplicative.

And the time available to sort through all these multiple options for complex maneuvers will be limited. MDO emphasizes the importance of rapidly exploiting windows of superiority, which appear unpredictably and last only a very short time. Exploiting such short-lived windows of superiority will often require rapid and major—potentially risky—modifications to ongoing plans, coordinated across multiple domains. AI can help orchestrate these changes, assess their ramifications, and produce, within seconds if necessary, the detailed fragmentary order required [to continue operations].”

This vision aptly summarizes the underlying challenge of integrating AI into command by intention: to make these tools multipliers of understanding and responsiveness in support of subsidiarity, without delegating ultimate responsibility for tactical decisions to the machine. 

The Case of the U.S. Maven Smart Systems Program: From ISR Analysis to the Decision-Making Platform

The United States has been among the pioneers in integrating conversational AI models into military systems. The platform proposed by Palantir in 2023 was described in the press as follows:

Palantir Technologies has unveiled its Artificial Intelligence Platform (AIP) to the public. At its core, this system integrates large language models (LLMs) and artificial intelligence capabilities within a secure private network. The armed forces can thus use AIP to develop battle plans almost as easily as students use ChatGPT to write an essay. ”

In a demonstration scenario described on March 13, 2026, in Wired magazine, an analyst interacts with a “digital assistant” integrated into this platform, which, after an artificial vision algorithm automatically detects abnormal activity, helps the analyst interpret the situation and consider several options: airstrike, long-range artillery, or deployment of a tactical team. The cognitive intermediary can also suggest approach routes or the allocation of electronic warfare assets, thereby accelerating the preparation for action.

These capabilities are part of the “Project Maven” ecosystem, the cornerstone of integrating artificial intelligence into U.S. military operations. These demonstrations are not only aimed at providing direct support for targeting but also serve as training tools to accustom staff and targeting cells to working with a “cognitive assistant” in wargames and planning exercises

By way of background: launched in 2017 by the Pentagon, Project Maven also known as AWCFT for “Algorithmic Warfare Cross-Functional Team” is the first large-scale initiative aimed at integrating artificial intelligence into intelligence analysis and targeting.

Originally, the program addressed a simple yet critical challenge: the explosion in the volume of images collected by drones and satellites far exceeded human analytical capacity. Maven was thus designed to automatically analyze these data streams, identify vehicles, infrastructure, or suspicious behavior, and assist analysts in designating potential targets.

The system is now managed by the National Geospatial-Intelligence Agency and accessible to all components of the Pentagon, which has become the “Department of War” (DoW). Over the years, it has evolved from an image analysis tool into a “broader military intelligence and targeting platform that fuses data from multiple sensors to identify objects, assess threats, and facilitate operational decision-making”

In concrete terms, the deployment of Maven Smart Systems is described by Alex Hollings as follows:

The Maven Smart System, already operational today, is capable of processing intelligence, surveillance, and reconnaissance (ISR) data streams in real time—including video, imagery, radar data, and radio signals—and then using computer vision based on convolutional neural networks to detect and classify people, vehicles, equipment, and much more, geolocating each while distinguishing between friendly forces, enemies, and civilians.

It can then use this data to propose up to 1,000 targets per hour to users, who can then turn to Maven’s AI-powered mission recommendation system to identify the most appropriate weapon to engage each target based on various factors, such as the type of munitions best suited to the mission, the platform’s flight time, weapon loadout details, and the location of friendly personnel and partner forces.

Maven was designed to address the challenge of rapidly processing data collected by intelligence, surveillance, and reconnaissance (ISR) aircraft and drones. Once the target is identified and the appropriate weapon and platform are determined, Maven can communicate directly with troops on the ground, or even directly with the platforms and weapon systems themselves. In 2020, Maven transmitted firing orders to a U.S. Army M142 HIMARS artillery system for the first time during tests at Fort Liberty.

By 2023, Maven had demonstrated its ability to interface directly with the Army’s mission command systems, such as the Advanced Field Artillery Tactical Data System (AFATDS), to generate fire missions in Qatar during actual combat operations as part of Operations Spartan Shield and Inherent Resolve.

In June 2025, Maven gained the capability to interface directly with the U.S. Army’s Air Mission Planning System (AMPS), thereby effectively automating the transition from conventional air mission planning systems to Maven’s Common Operational Picture (COP), and creating a one-stop shop for effective flight mission planning using the latest available intelligence.

By January 2026, Maven had been deployed across all major U.S. combat commands, as well as in operations of NATO allied commands.”

The integration of conversational models into this environment represents, in a sense, the third stage of AI integration in the defense sector. After the analysis phase (2017) and the AIP technology demonstrator (2023), the operational implementation phase has been proceeding gradually since 2024: analysts can now interact with data using natural language, request information summaries, or rapidly generate multiple tactical scenarios.

Accelerating the OODA loop: Toward decision-making superiority

While the Maven program is currently one of the most successful examples of this transformation, it should not be viewed in isolation, but rather as one element within a wider ecosystem of systems and architectures—often grouped in the United States under the banner “Joint All-Domain Command and  Control” (JADC2) aimed at integrating artificial intelligence into all aspects of operations.

Within the U.S. Air Force, programs such as Skyborg aim to integrate autonomous decision-making capabilities directly at the effector level and into a drone accompanying a piloted aircraft (“loyal wingman”). For their part, ABMS (“Advanced Battle Management System”) architectures which constitute the U.S. Air Force’s contribution to the JADC2 program seek to connect sensors, platforms, and command centers within a multi-domain combat framework, leveraging AI and machine learning algorithms to fuse data and drastically reduce decision-making times.

These tools enable the rapid analysis of large volumes of data from radar or satellite imagery, sensors, or intelligence reports to identify suspicious enemy activities, synthesize the tactical situation, and propose multiple tactical options. The goal is not to replace human decision-making, but to accelerate the planning process by drastically reducing the time between detection, analysis, and action. This objective aligns directly with the principle of Command by Intention, according to which, while command sets the objective and intention, subordinate units make tactical decisions at the appropriate momen, a principle that is all the more relevant today in a context of high-intensity, multi-domain operations and communications jamming, where it is becoming increasingly difficult to centralize everything.

AI effectively bridges technology with a classic strategic concept: this evolution is essentially an AI-assisted version of the OODA loop. Briefly, the OODA loop — “Observe, Orient, Decide, Act” —was conceptualized by U.S. Air Force Colonel John Boyd and boils down to the following idea: in a conflict, the advantage goes to whoever can cycle through this process faster than the adversary. However, if we look specifically at the step-by-step AIP demonstration, we find almost exactly this logic:

• Observe: the system detects abnormal activity through automated analysis of sensor data. This function corresponds to capabilities developed over several years as part of the Project Maven program, which aims to automate the analysis of imagery and intelligence streams.
• Orient: Once the activity is detected, the AI helps the analyst understand the situation. For example, the system proposes a hypothesis about the nature of the enemy unit by cross-referencing various clues. This is the orientation phase of the OODA loop: transforming raw data into situational awareness.
• Decide: The analyst then asks the system to generate several tactical options. The AI proposes various possible courses of action (airstrikes, long-range artillery, ground unit operations, etc.). The final decision remains human, but the AI accelerates the generation of options.
• Act: Finally, the system can assist in implementing the chosen option: generating approach routes, allocating resources, or coordinating with other capabilities such as electronic warfare.

AI does not replace the OODA loop, but seeks to accelerate it.

By automating certain tasks, detection, data fusion, and option generation, the system reduces the time required for a human to move from observation to action, conferring a significant tactical and decision-making advantage.

Bibliography

Bellanger, Jérôme, and Tsiporah Fried. 2026. “Command, Speed, and Aerospace Superiority in the Era of High-Intensity Conflicts.”  Revue Défense Nationale,  887 (February, 2026).

Center for Strategic and International Studies. n.d. “Calibrating NATO’s Vision for AI-Enabled Decision Support.”

Clement. 2024. “NATO and AI.” NATO Parliamentary Assembly. https://www.nato-pa.int/document/2024-nato-and-ai-report-clement-058-stc.

French Air and Space Force / CESA. 2024. “French ACE Operational Concept: The French Adaptation of Agile Combat Employment.” December. https://www.defense.gouv.fr/cesa.

French Ministry of the Armed Forces. n.d. “Commandement par Intention.” In Vision stratégique du chef d’état-major de l’Armée de terre. https://www.defense.gouv.fr/terre/chef-detat-major-larmee-terre/vision-strategique-du-chef-detat-major-larmee-terre/commandement-intention-0.

Hollings, Alex. n.d. “The Maven Smart System.” Sandboxx News. https://www.sandboxx.us/news/maven/.

Marine Corps University Press. n.d. “Artificial Intelligence-Enabled Military Decision-Making Process.” Journal of Advanced Military Studies , 16 (2). https://www.usmcu.edu/Outreach/Marine-Corps-University-Press/MCU-Journal/JAMS-vol-16-no-2

Metz, Cade. 2026. “Palantir Demos Show How the Military Can Use AI Chatbots to Generate War Plans.” Wired, March 13. https://www.wired.com/story/palantir-demos-show-how-the-military-can-use-ai-chatbots-to-generate-war-plans/.

Modern War Institute at West Point. 2022. “Beyond the Hype: Why We’re Closer to AI-Enabled Mission Command than You Think.” Modern War Institute. https://mwi.westpoint.edu/beyond-the-hype-why-were-closer-to-ai-enabled-mission-command-than-you-think/.

National Academies of Sciences, Engineering, and Medicine. 2022. “Command and Control in Joint All-Domain Operations.” In Testing, Evaluating, and Assessing AI-Enabled Systems for the Department of Defense. https://www.nationalacademies.org/read/26525/chapter/2#2.

NATO. 2024. “Summary of NATO’s Revised Artificial Intelligence (AI) Strategy.” July 10. https://www.nato.int/fr/about-us/official-texts-and-resources/official-texts/2024/07/10/summary-of-natos-revised-artificial-intelligence-ai-strategy.

Palantir Technologies. n.d. “Maven Smart System: Innovating for the Alliance.” Palantir Blog. https://blog.palantir.com/maven-smart-system-innovating-for-the-alliance-5ebc31709eea.

Palantir Technologies. n.d. “Palantir Announces Artificial Intelligence Platform for Enterprise and Military Use.” Maginative. https://www.maginative.com/article/palantir-announces-artificial-intelligence-platform-for-enterprise-and-military-use.

Palantir Technologies. n.d. “Maven Smart System – Innovating for the Alliance.” YouTube video featuring Cameron Stanley. https://www.youtube.com/watch?v=yrtDgoqWmgM.

Pardue, Jeff, et al. 2025. “The U.S. Army, Artificial Intelligence, and Mission Command.” War on the Rocks, March. https://warontherocks.com/2025/03/the-u-s-army-artificial-intelligence-and-mission-command/.

Project Flux. n.d. “Palantir’s Maven Smart System Revolutionises Pentagon Decision-Making.” Project Flux. https://www.projectflux.ai/p/palantir-s-maven-smart-system-revolutionises-pentagon-decision-making.

Snow, Shawn. 2026. “Pentagon Expands Palantir’s Role in AI Contract.” Military.com, March 22. https://www.military.com/feature/2026/03/22/pentagon-expands-palantirs-role-ai-contract.html.

U.S. Air Force. 2021. “Skyborg: Rise of the Autonomous Wingmen.” Airman Magazine. https://www.airmanmagazine.af.mil/Features/Display/Article/2604028/skyborg-rise-of-the-autonomous-wingmen/.

U.S. Air Force Life Cycle Management Center. 2024. “Understanding Advanced Battle Management System (ABMS).” Podcast. https://www.aflcmc.af.mil/NEWS/Article-Display/Article/3792745/understanding-advanced-battle-management-system-abms-podcast/.

U.S. Department of War. 2017. “Project Maven Industry Day Pursues Artificial Intelligence for DoD Challenges.” October. https://www.war.gov/News/News-Stories/Article/Article/1356172/project-maven-industry-day-pursues-artificial-intelligence-for-dod-challenges.

U.S. Marine Corps University / War on the Rocks. 2025. “Human-Machine Planning: AI Lessons from the Marine Command and Staff College.” War on the Rocks, October. https://warontherocks.com/2025/10/human-machine-planning-ai-lessons-from-the-marine-command-and-general-staff-college/.

U.S. Army. n.d. “Innovating Defense: Generative AI’s Role in Military Evolution.” U.S. Army., https://www.army.mil/article/286707/innovating_defense_generative_ais_role_in_military_evolution

Airforce-Technology.com.  “Uncrewed Ambitions of the Loyal Wingman.” Airforce-Technology. https://www.airforce-technology.com/features/uncrewed-ambitions-of-the-loyal-wingman/.

Conversational AI and Intent Based Command: Training at the New Frontier of Command