Securing America’s Decision Advantage

For the last two years, the conversation around AI in national security has been dominated by Large Language Models. These tools, while impressive, have largely functioned as clever junior staffers, summarizing emails or drafting memos, yet unable to execute complex tasks or interact with the physical world.

War, however, is not fought with text. The future of military AI is not larger models alone, but collaborative networks of heterogeneous AI agents.

Today, Scale AI is releasing a plan for how the Department of War can deploy systems of AI Agents to revolutionize military planning and command and control. "The Agentic Revolution in War: The Present and Future of Decision Advantage," is available for download at scale.com/agentic-warfare.

While the Department of War has made steps to expand the use of its AI arsenal, the time has come to move to true agentic systems, or constellations of AI agents that can monitor the battleplace, plan, test, and execute complex actions at speeds no human staff can match.

“The Agentic Revolution in War” provides a blueprint for how the Department can turn the prototype systems of today into a permanent part of its arsenal, providing enduring strategic superiority and a new means of deterrence by decision advantage. 

Problem: Fighting a 2026 War with an 1803 Process

The United States possesses the world’s most advanced sensors and shooters. Yet, the process connecting them, commonly called the "OODA Loop" (Observe, Orient, Decide, Act), is breaking under the weight of data.

As we detail in the white paper, the U.S. military’s current planning process hasn't fundamentally changed since Napoleon codified the general staff system in 1803. We still rely on linear, manual workflows that produce static Operational Plans in physical binders that take two years to write and are often obsolete by the time they are printed.

In a conflict with a near-peer adversary like China, we will not have two years; we may not even have two days.

Solution: Decisional Advantage

Scale is building an AI-agent based architecture to solve this. We define Agentic Warfare as deterrence by decisional advantage.

This advantage does not come from a single, massive "black box" model. It comes from teams of heterogeneous agents: specialized workers that use different tools and reasoning engines to solve specific parts of a problem. Imagine a game of chess where one player can make three moves for every one move their opponent makes. That is what agentic systems offer. By compressing the time required to analyze data and generate options, these systems allow U.S. commanders to out-think and out-sequence the adversary to make informed decisions.

Our white paper details two specific architectures Scale is developing to apply this leverage:

1. Observe with Agentic Alerting: In a saturated battlespace, human operators are blinded by noise. Agentic alerting systems run on the edge, using cross-cueing to autonomously verify threats. When a low-fidelity sensor picks up an anomaly, it instantly tasks a high-fidelity sensor to investigate. This turns a reactive force into a proactive one, flagging the signal in the noise before the threat materializes.

2. Decide with Agentic Planning: This is the death of the static binder. We are pioneering systems through our work on Thunderforge that allow planners to generate a planning multiverse. Agents will be designed to run thousands of physics-based simulations overnight to stress-test assumptions and validate Courses of Action (COAs). This ensures that when a commander makes a decision, it is backed by probability-weighted analysis, creating "living plans" that self-heal as the battlefield changes.

From "In the Loop" to "On the Loop"

A key theme of our paper is the changing role of the human commander.

The speed of modern conflict requires us to move commanders from being "in the loop" and manually reviewing every data packet (which creates a fatal bottleneck) to being "on the loop" and orchestrating the machine’s logic and setting strategic intent.

This elevates the human from a data processor to a true strategist, far from surrendering human control.

Trust is the Prerequisite

We also know that speed is useless without trust. That is why our white paper also introduces Scale’s Horizon-Gate Framework, a taxonomy for AI delivery. We differentiate between research-gated capabilities, which are promising but unproven, and engineering-gated capabilities that are robust, accredited, and mission-ready. Our research pioneers the rigorous Test & Evaluation regimes necessary to ensure that agentic systems respect rules of engagement and policy envelopes.  

The Window is Closing

The technology to achieve this advantage exists today, but the window to secure it is closing. Adversaries including China are already reorganizing their forces around so-called “intelligentized warfare" and "command brains" designed to cognitively overwhelm adversaries who are not using AI-enabled systems and collapse American decision-making. We no longer need to ask if AI will revolutionize command and control; the question is whether the United States is prepared to lead that transformation.

Read our full blueprint for the Agentic Era at scale.com/agentic-warfare.