{"slug": "borrow-what-works-the-case-for-army-adoption-of-ukraines-combat-tested-kill-web", "title": "Borrow What Works: The Case for Army Adoption of Ukraine’s Combat-Tested Horizontal Kill Web Architecture", "summary": "The U.S. Army should adopt Ukraine's combat-tested horizontal kill web architecture, which uses the DELTA battlefield management system, Vezha live-video dashboard, and AI-based Avengers target-detection platform to shorten sensor-to-shooter timelines, according to an observer with Security Assistance Group–Ukraine. The system enables authenticated users at different echelons to share real-time observations, with AI providing four layers of assistance from general-purpose tools to platform-level cues.", "body_md": "Thirty live video feeds fill an operator’s [Vezha dashboard](https://www.pravda.com.ua/eng/news/2024/10/14/7479592/). More are available, but these are the thirty selected for the current fight. The operator has filtered the larger pool by unit and geography: reconnaissance crews supporting the maneuver, adjacent teams whose sensors overlap the area of operations, and other feeds likely to affect the mission. The operator knows where those crews are working and can watch what they are seeing in real time.\n\nAn alert flashes beside one of the feeds. The AI-based [Avengers platform](https://thedefensepost.com/2026/08/12/ukraine-ai-drone-target-detection/) has detected a probable vehicle in another crew’s video. A human reviews the cue, classifies the object, and publishes it to a [DELTA layer](https://cepa.org/article/the-heart-of-war-ukraines-key-battlefield-system/). The record appears with coordinates, time, imagery, source, and the reporting user’s handle. The vehicle is inside the unit’s area. Another asset can engage it, and the local commander has already delegated the authority to act. If more detail is needed, the crew that created the record can be contacted directly through secure chat.\n\nThe same object may also be reviewed by 2IIC, an intelligence integration cell that supports the Ukrainian Unmanned Systems Forces and curates and distributes shared layers. Local crews continue adding observations while analysts refine the classification, movement, or history. Every authorized user sees the update on the same object. No one has to rebuild the packet, retype the grid, or wait for a screenshot to pass through several headquarters. The sensor may belong to an adjacent unit, the analyst may work elsewhere, and the available shooter may belong to the unit that owns the area. This is the horizontal kill web, and I had the chance to observe its elements during a recent assignment with Security Assistance Group–Ukraine.\n\nThe US Army usually describes the same process as a sensor-to-shooter chain. A sensor reports upward. An intelligence section validates the report. A headquarters prioritizes it. An operations or fires cell assigns an effector, and the tasking moves back down. Every handoff adds time and another chance to lose context. A vehicle can leave the open, enter cover, or move outside the assigned weapon’s range while the report is still traveling.\n\nDELTA shortens that path. Authenticated users at different echelons work in the same browser-based environment. Vezha (Вежа, or “tower”) organizes live video and allows each user to select the streams relevant to the mission. The Avengers platform continuously analyzes connected video, primarily for vehicles and military equipment. Once a human validates a cue and adds it to a layer, the observation [becomes available](https://www.csis.org/analysis/does-ukraine-already-have-functional-cjadc2-technology) to the crews, analysts, commanders, and effectors with permission to use it.\n\n**Four Layers of AI in the Fight**\n\nArtificial intelligence in Ukraine already reaches well beyond drones and targeting applications. The broadest layer is the same general-purpose AI now used across civilian workplaces. Soldiers use AI tools to translate, draft and edit reports, summarize long documents, compare data, research unfamiliar systems, write software, prepare briefings, and develop white papers. These tasks are ordinary, but doing them faster affects the tempo of a headquarters and the quality of information reaching the tactical edge.\n\nThe next layer is platform-level assistance. In drone footage I had the opportunity to review, some systems place boxes and probable classifications directly on the operator’s live display. A crew can receive a real-time cue for a vehicle, person, or other object while the drone is still flying. The processing may occur onboard or in the associated control station, depending on the system. Either way, the cue helps the crew search its own field of view and decide where to look more closely.\n\nNetwork-level AI works across feeds rather than inside one aircraft. Vezha can make a much larger pool of live video available than any operator displays at once. Users select feeds by unit, location, and mission while the Avengers Labs platform inspects connected video in the background. Ukraine’s Ministry of Defence [reports that Avengers](https://mod.gov.ua/en/news/enemy-equipment-detected-in-2-seconds-the-ministry-of-defence-showcased-delta-and-avengers-systems-at-the-london-defence-conference) identifies 70 percent of enemy vehicles and equipment visible in processed video and detects an individual item in about 2.2 seconds. The practical result is a persistent first look across video that would otherwise depend on humans staring at screens for hours.\n\nThe fourth layer is terminal or last-mile assistance. After a human selects a target or aim point, computer vision can maintain the track during the final approach. Pixel lock and related [visual-tracking functions](https://auterion.com/auterion-secures-contract-to-deliver-33000-skynode-drone-strike-kits-to-ukraine/) are especially useful when electronic warfare degrades the control or video link. They reduce the amount of correction required from the pilot and can carry an attack through the last seconds of flight.\n\nThese layers do not have to be packaged together. A drone may show local detections without connecting to DELTA. A feed may enter Vezha from an aircraft with no onboard AI. A strike drone may use terminal tracking against a target found by another sensor. Most current autonomy is narrow and task-specific. End-to-end systems that independently search, select, and engage targets remain uncommon. Ukraine’s advantage comes from overlapping forms of assistance that reduce workload and allow information to continue moving.\n\n**From a Video Cue to a Living DELTA Layer**\n\nA common operational picture is often imagined as a map covered with symbols. DELTA’s greater value is the record behind each symbol. A DELTA object can retain an image or video, classification, affiliation when known, location, timestamp, source, previous observations, current status, and the name or handle of the person or crew that reported it. A user who needs clarification can open the supporting evidence and contact the source directly.\n\nThe layers are living intelligence products. Local crews add observations as they occur. Intelligence organizations such as 2IIC can combine those reports with other analysis, correct or refine entries, and distribute the layer across the force. Users can filter the picture by unit, geography, target type, status, time, and operational depth. A commander preparing to maneuver through one sector can focus on enemy vehicles, artillery, electronic warfare systems, air defenses, obstacles, friendly drone activity, or any other category relevant to the mission.\n\nThe first observation is rarely complete or permanent. A vehicle may disappear under trees. A suspected position may be seen from several angles. An electronic warfare team may recognize an emitter that a video analyst cannot. Another reconnaissance crew may find the same equipment later in a different location. When those observations remain attached to the same object, each user inherits the work already completed. The target history becomes more useful with every valid update.\n\nThe Vezha workflow shows how quickly that can begin. AI detects a probable object in a stream. A human classifies it only to the level the evidence supports: perhaps “car,” “howitzer,” “personnel,” friendly, enemy, or unknown. The user then publishes it to the appropriate DELTA layer. Users operating in or near that location can see the object immediately and can receive an alert when the system is configured to notify them. One geospatial record replaces separate reports sent to every organization.\n\nDELTA also reaches users without tying them to one specialized workstation. Phones, tablets, and computers can display the same relevant objects through a web browser or mobile application. Access is tied to the individual and the mission. [Public descriptions of the system](https://www.csis.org/analysis/does-ukraine-already-have-functional-cjadc2-technology) identify role-based permissions, strong authentication, managed applications, auditability, and the ability to revoke or wipe access. In practice, users may authenticate through a hardware security key, a phone-based second factor, or another approved method. Wide access does not mean anonymous access or universal access to every layer.\n\n**The Closest US Equivalent Is a Stack of Systems**\n\nThe Army already owns capable pieces of this problem. [ATAK and the wider TAK ecosystem](https://tak.gov/products) share positions, overlays, messages, and selected data at the tactical edge. The [Command Post Computing Environment](https://ac.devcom.army.mil/wp-content/uploads/sites/5/2025/02/CommandPostComputingEnvironment.pdf) (CPCE) supports the command-post common operational picture. The [Advanced Field Artillery Tactical Data System](https://www.army.mil/article/283097/afatds_gets_an_upgrade) (AFATDS) processes fires. [Maven Smart System](https://www.csis.org/analysis/what-maven-smart-system-and-what-does-it-do) fuses and analyzes large data sets and supports AI-enabled targeting. Drone crews also use platform-specific ground-control stations and video viewers.\n\nA target can cross several of those systems before a weapon is assigned. The drone crew may watch the live feed on its ground-control station, report the grid by voice or chat, and send a screenshot to a staff section. The target may be entered into CPCE or another mission-command application, developed in an intelligence system, and then typed into AFATDS for a fire mission. Different units and partners may be working on separate networks. When information is placed on SIPRNet or another classified enclave, it may exclude the company commander, coalition partner, or drone crew that needs the result but lacks the terminal, account, or releasability.\n\nSoldiers bridge these seams every day. They use voice calls, screenshots, chat rooms, spreadsheet trackers, and manually reentered coordinates. Those methods can move a target, but they demand attention and create delay. Each application may perform its own function well while the overall process still depends on a soldier copying data from one place to another.\n\nMaven Smart System is the [closest American analogue](https://www.reuters.com/technology/pentagon-adopt-palantir-ai-as-core-us-military-system-memo-says-2026-03-20/) in data fusion and AI-assisted targeting. Yet it remains oriented toward a different architecture and, in the author’s experience, has not provided the same intuitive, whole-force tactical environment that DELTA places in the hands of local crews. The relevant measure is whether a reconnaissance crew, intelligence analyst, local commander, and shooter can work from the same live object without rebuilding it.\n\n**Why Horizontal Is Faster**\n\nOnce a target is published to a shared layer, several functions can occur at the same time. A reconnaissance crew can maintain observation. An analyst can refine identification. An electronic warfare team can add emitter information. A strike crew can determine whether it has the range, payload, and time to act. A commander can change the priority. All of these actions can occur in parallel.\n\nCommanders still establish priorities, boundaries, fire-control measures, restrictions, and engagement authority. Ukrainian practice pushes much of that authority to the lowest level able to act. The commander who owns the area can engage a valid target with an available weapon instead of returning each decision to a distant headquarters. Higher echelons retain visibility and can redirect effort, but they do not have to serve as the relay for every observation and every question.\n\nDirect contact with the source also preserves context. A strike crew can ask whether the vehicle is still moving. An analyst can explain why a classification changed. A commander can determine how current the observation is before assigning a scarce weapon. The answers come from the people who saw or developed the target, not from a summary several reporting steps removed from the event.\n\nBrowser-based access lets this happen across echelons. A brigade staff, battalion commander, drone crew, electronic warfare team, and adjacent unit can receive different views of the same operational picture according to their permissions. The security model follows the authenticated user and the data instead of relying entirely on physical access to one type of terminal. Strong individual authentication and rapid revocation can provide high identity assurance while keeping time-sensitive information available at the tactical edge.\n\n**Ukraine’s Organizational Lessons for the US Army**\n\nA recent rotation at the Joint Multinational Readiness Center in Germany placed Ukrainian unmanned systems practitioners against a US armored formation. A Ukrainian platoon-sized element [stopped an armored brigade](https://www.wsj.com/politics/national-security/u-s-and-ukrainian-forces-went-head-to-head-in-an-exercise-ukraines-drones-won-cc3663d5) in its tracks. The result was not surprising to people familiar with the problem set. The Ukrainian teams brought persistent reconnaissance, specialized strike and electronic warfare crews, DELTA, shared layers, direct horizontal communication, and authorities already pushed to the people executing the fight.\n\nThe US formation had more mass and far more conventional combat power. It could not integrate its sensors, staffs, and available effects at the same speed. American crews often saw pieces of the fight, but the information did not move through the formation quickly enough to create a comparable response. Ukrainian reconnaissance crews developed targets that strike crews could use immediately. Their synchronizers stayed connected to the crews and the live picture instead of waiting for a formal update cycle.\n\nThe lesson is organizational. The Army often trains a drone operator, evaluates flight proficiency, and declares the capability fielded. Ukrainian practice develops crews and connects them to other crews. The pilot, navigator or observer, analyst, engineer, electronic warfare support, battle captain, intelligence cell, and available effector operate as one system.\n\nInstructors at the Army’s Air Cavalry Leaders Course and the Unmanned Advanced Lethality Course routinely ask students how much experience they have in large-scale combat operations. The answer is almost always none. Ukraine has accumulated more than four years of continuous large-scale combat experience with DELTA and the surrounding unmanned systems ecosystem. No American [flyoff](https://drone-dominance.io/announcements.html?post=gauntlet-ii-advancing-companies), laboratory test, or short combat training center rotation can reproduce that record.\n\n**Last-Mile Autonomy Depends on the First Mile**\n\nTerminal tracking can help an attack survive a degraded link, but it depends on the quality of the information that reached the shooter. A stale coordinate, weak identification, or poor handoff can send a technically successful drone to the wrong place. Platform-level detection, network-level triage, the shared DELTA object, and target history improve the first part of the engagement before pixel lock is ever activated.\n\nCurrent last-mile systems also remain less capable than many presentations suggest. Experienced Ukrainian pilots frequently choose to fly the terminal attack manually because they achieve a higher probability of kill than they do with available visual-tracking modes. A skilled pilot can adjust the exact aim point, respond to target movement, handle partial concealment, and recognize when the automated track has attached to the wrong feature. Pixel lock is useful when the link begins to fail or the operator cannot maintain control, but it is not yet the preferred choice for many experienced crews.\n\nThat balance [will change](https://en.defence-ua.com/weapon_and_tech/whats_special_about_the_hornet_uav_that_ukrainians_using_to_destroy_russian_logistics_where_did_it_come_from_and_what_are_its_key_features-18597.html) as processors, models, sensors, and training data improve. The near-term requirement is still practical assistance: stabilization, object persistence, reacquisition after a short interruption, route following, obstacle avoidance, target handoff, and terminal tracking. The common operational environment should record how those systems perform so that battlefield results can improve the next model and the next mission.\n\n**Buy the Architecture That Already Works**\n\nThe Army should pursue a licensing and fielding arrangement for DELTA now. Four years of use in large-scale combat provide a test record that no domestic program can match. DELTA has operated under persistent surveillance, electronic warfare, cyberattack, attrition, rapid force adaptation, and coalition support. It has been used by units that must produce results every day rather than prepare for a demonstration.\n\nThe details would require negotiation. The United States would need appropriate hosting, cybersecurity review, data rights, support arrangements, interfaces with US systems, and procedures for American and coalition information. But the work should begin with the existing system. Ukraine has already built and refined the core system under the most demanding available standard.\n\nOther countries are paying attention. Ukrainian officials report international interest in DELTA, and the system [has exchanged data](https://www.csis.org/analysis/does-ukraine-already-have-functional-cjadc2-technology) with Poland’s TOPAZ artillery command-and-control system and participated in NATO interoperability events. The Army should treat DELTA as an operational baseline while continuing to connect Maven, TAK, CPCE, AFATDS, aviation mission systems, intelligence tools, and future sensors and effectors.\n\nAcquisition habits often favor a new domestic program, a new prime contract, and a new set of requirements. That approach consumes years while units continue training on disconnected systems. The Army can protect American data, negotiate sovereign control, and support domestic integration without discarding a battle-tested starting point. The soldiers who will depend on the system should not bear the risk created by institutional pride or the desire to reinvent a proven design.\n\nFielding the software is only one part of the work. The Army must train the whole kill web: reconnaissance crews, analysts, intelligence integration cells, electronic warfare teams, fires personnel, commanders, signal support, and the units that own the area. Exercises should measure the time from detection to a correct engagement decision, the number of manual data transfers, the quality of target history, and the ability to continue when links degrade. Authorities should be established before contact so that local commanders can act when a target appears.\n\nThe system must also be tested against real failure. Units need to train with congested networks, stale tracks, duplicate reports, conflicting classifications, camouflage, decoys, compromised accounts, and interrupted cloud access. Local caching and disconnected procedures should preserve the picture long enough for the fight to continue and synchronize again when communications return.\n\nFinally, integration should become a threshold requirement in the Army’s drone competitions. Range, speed, endurance, payload, cost, and resistance to electronic warfare remain important. Yet a drone that flies another kilometer or travels ten kilometers per hour faster may still produce less combat value if its video, metadata, detections, and tasking cannot move through the common operational environment. The preferred system should publish to the shared layer, accept tasking from it, preserve source and target history, and fit the crew workflow already used by the formation.\n\nThe horizontal kill web is an architecture for allowing sensors, AI tools, intelligence organizations, local commanders, and effectors to work from the same live record. For Ukraine, DELTA provides the shared environment and layers. Vezha organizes the video selected for each mission. The Avengers platform gives human reviewers a persistent first look at probable vehicles and equipment. Platform-level AI helps crews search their own feeds. General-purpose AI accelerates the routine work around the fight. Terminal tracking supports the final approach when the link degrades.\n\nUS military [drone competitions](https://drone-dominance.io/announcements.html?post=gauntlet-ii-advancing-companies) can identify useful aircraft, but the best architecture will determine how much combat power those aircraft deliver. Marginal improvements in range or speed should not outweigh the ability to join a shared operational picture, move targeting data without manual reentry, and connect an adjacent sensor to the local commander and shooter. Ukraine has already demonstrated that architecture in large-scale combat. The United States should field it, train with it, and build future drones to operate inside it.\n\n*Chief Warrant Officer 3 James L. Andreasen is an AH-64E Apache instructor pilot and aviation mission survivability officer at Fort Rucker, Alabama, where he serves as an instructor in the Air Cavalry Leaders Course. Following a six-month assignment with Security Assistance Group–Ukraine, he contributes to development of the Unmanned Advanced Lethality Course and supports Army aviation modernization efforts focused on unmanned systems integration.*\n\n*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.*\n\nImage credit: [Ministry of Defence of Ukraine](https://mod.gov.ua/en/news/the-delta-combat-system-has-been-deployed-across-all-levels-of-defence-forces-of-ukraine) (Creative Commons Attribution 4.0 International license)", "url": "https://wpnews.pro/news/borrow-what-works-the-case-for-army-adoption-of-ukraines-combat-tested-kill-web", "canonical_source": "https://mwi.westpoint.edu/borrow-what-works-the-case-for-army-adoption-of-ukraines-combat-tested-horizontal-kill-web-architecture/", "published_at": "2026-08-14 05:05:39+00:00", "updated_at": "2026-08-14 05:14:14.349753+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-tools", "ai-infrastructure"], "entities": ["U.S. Army", "Security Assistance Group–Ukraine", "DELTA", "Vezha", "Avengers", "Ukrainian Unmanned Systems Forces", "2IIC"], "alternates": {"html": "https://wpnews.pro/news/borrow-what-works-the-case-for-army-adoption-of-ukraines-combat-tested-kill-web", "markdown": "https://wpnews.pro/news/borrow-what-works-the-case-for-army-adoption-of-ukraines-combat-tested-kill-web.md", "text": "https://wpnews.pro/news/borrow-what-works-the-case-for-army-adoption-of-ukraines-combat-tested-kill-web.txt", "jsonld": "https://wpnews.pro/news/borrow-what-works-the-case-for-army-adoption-of-ukraines-combat-tested-kill-web.jsonld"}}