The U.S. Department of Defense has launched Agent Network, an agentic AI tool that continuously scans defense intelligence and operational networks, translates findings into targeting options, and presents them to military commanders “within seconds,” according to Defense One’s Patrick Tucker. The system explicitly prevents autonomous target selection or strikes. Commanders retain decision authority over every targeting action.
How Agent Network Works
Agent Network uses autonomous AI agents to search intelligence feeds, correlate data across operational systems, and generate option sets for commanders. The Pentagon described the system in a press release: agents “continuously scan defense intelligence and operational systems, translating findings into clearly presented options.” The agents execute tasks autonomously (running searches, cross-referencing data, building option sets) while the final targeting decision stays with human commanders.
Key contractors on the project include Lumbra and Palantir, which already handles targeting analysis through its Maven Smart Systems contract, Defense One reported. Agent Network is one of seven “pace-setting” projects unveiled in January alongside the Pentagon’s new AI strategy.
The Capability Debate
Whether current large language models can reliably handle the computational demands of real-time battlefield decision support is contested. Vishal Sikka, former CEO of Infosys, argued last July that transformer-based models face fundamental limitations when task complexity exceeds their token processing capacity. Citing the Time-Hierarchy Theorem, Sikka wrote that “extreme care must be used before applying LLMs to problems or use cases that require accuracy, or solving problems of non-trivial complexity,” as quoted by Defense One.
Illia Pashkov, founder of SINT Labs and editor of The Agent Times, pushed back. “Agentic AI quietly stopped being a demo this year,” Pashkov told Defense One. “It’s drafting code, clearing support queues, grinding through back-office work in finance and healthcare, and now it’s reading intelligence. The speed is not hype. I’ve watched these systems compress weeks of analyst work into an afternoon.”
The Governance Problem
Pashkov also identified the core risk: agent confidence without oversight. “The danger was never a dumb agent; it’s a confident one running without a leash, a logbook, or a human who owns the call,” he told Defense One.
That concern is already surfacing inside the Department of Defense. One DOD intelligence security official, not directly affiliated with Agent Network, described an atmosphere of enthusiasm about agent deployments across multiple offices and teams. “There are so many opportunities to leverage the DOD Enterprise capabilities and allow people to build their own agents,” the official said. But the official acknowledged that keeping track of how every agent performs is a major challenge, and governing all of them will be “nearly impossible.”
Decision Velocity and Escalation Risk
Agent Network compresses the intelligence-to-options pipeline from hours or days to seconds. That speed is the stated goal. It is also the risk. When a system generates targeting options faster than a commander can evaluate the underlying intelligence, the pressure to act on agent-generated recommendations increases. The formal guardrail (commanders retain decision authority) depends on commanders having enough time and context to meaningfully exercise that authority rather than rubber-stamp agent output.
The Pentagon has not disclosed how Agent Network handles edge cases: conflicting intelligence, low-confidence assessments, or scenarios where the agent’s option set excludes viable alternatives that a human analyst would have identified. These are the failure modes that matter most in targeting decisions, and the ones that current LLM architectures handle least reliably.