The Rise of Agentic AI in Defense: A Veteran's Perspective

The term "AI" has been thrown around the Pentagon and tech startups alike for years. But what we're seeing now—specifically with the rise of Agentic AI—is not just an incremental improvement. It's a fundamental shift in how software operates, and for those of us with a military background, it feels eerily familiar.

From Tools to Teammates

Traditional software is a tool. You give it an input, it gives you an output. It's a rifle: precise, effective, but inert without an operator. Generative AI (like the early GPT models) was a better tool—maybe a smart scope that calculates windage and elevation for you.

Agentic AI is different. It's not a rifle; it's a squad member. You give it a mission ("Secure this network perimeter," "Optimize this supply chain for a contested environment"), and it figures out the steps to get there. It observes, orients, decides, and acts. It loops. It learns.

"The difference between a tool and an agent is agency. A tool waits for a command. An agent pursues a goal."

The OODA Loop at Machine Speed

Every veteran knows Boyd's OODA Loop: Observe, Orient, Decide, Act. In combat, the side that executes this loop faster wins. Agentic AI systems are essentially OODA loops instantiated in code.

1. Observe

Agents ingest massive amounts of telemetry data—server logs, satellite imagery, network traffic patterns—far beyond human capacity.

2. Orient

They contextualize this data against their mission parameters. "Is this traffic spike a DDoS attack or just the Monday morning login rush?"

3. Decide

Based on pre-defined Rules of Engagement (ROE), they formulate a plan. This is where the "Agentic" part shines. They can weigh multiple courses of action and select the optimal one.

4. Act

They execute the code, patch the vulnerability, or reroute the supply convoy. And then they immediately go back to Step 1 to observe the effects of their action.

The "Human-on-the-Loop" Necessity

This autonomy is powerful, but it's also where the risk lies. In the military, we have the concept of "Command and Control" (C2). You delegate authority, not responsibility. The same must apply to Agentic AI.

At Defendre Solutions, we advocate for a "Human-on-the-Loop" architecture for critical systems, rather than "Human-in-the-Loop."

• Human-in-the-Loop: The AI pauses and waits for approval before every action. (Too slow for cyber warfare).
• Human-on-the-Loop: The AI acts autonomously within set boundaries, but a human supervisor monitors the operation and can intervene or "abort" at any moment. (The sweet spot).
• Human-out-of-the-Loop: The AI operates with no oversight. (Unacceptable for lethal or critical infrastructure systems).

Why Veterans Are Uniquely Suited for This Era

It's no coincidence that many of the leading voices in responsible AI defense have military backgrounds. We understand Rules of Engagement. We understand Chain of Command. We understand that a system that acts without accountability is a liability, no matter how "smart" it is.

Developing Agentic AI isn't just a coding challenge; it's a doctrine challenge. It requires defining:

1. Left and Right Limits: What is the agent absolutely never allowed to do?
2. Escalation Protocols: When must the agent wake up a human?
3. Fail-Safe Mechanisms: How do we "kill" the agent if it goes rogue or gets compromised?

The Future is Autonomous, But Accountable

We are building these systems today. From automated server defense bots that patch zero-days in milliseconds, to logistics agents that reroute supplies around weather and threats autonomously. The technology is here.

The mission now is to build the governance to match the capability. That's the Defendre Solutions approach: cutting-edge autonomy, wrapped in military-grade discipline and accountability.

Because in the end, an agent is only as good as the doctrine it follows.