This is what I’m working on today, the shift as large language models (LLMs) evolve from predictors into systems layered with new agentic capabilities. Once planning, reflection, and tools are wrapped around the model, AI stops being just a helper at the edges of staff work and starts behaving like a system actor. The models have already shifted, but have we.
To show how quickly this cascades, I’ve mapped it across six flows, each one tracing how the layering shift moves outward: from workflows into force design, from doctrine into command, and from coalition practice into policy for defence.
Flow 1: The Disruption
The evolution of LLMs is disrupting at the core, pushing militaries to confront whether they are ready to operate with machine-mediated orchestration at the core of command, intelligence, and coalition coordination.
The shift is no longer about speeding up staff work. It is about reorganizing the foundations of how defence institutions design, command, and coordinate force.
Flow 2: Force Design
AI becomes less about individual tools (chatbots, copilots) and more about system architecture. The critical questions ahead:
Who owns the orchestration layer in joint operations?
How are agent workflows certified for coalition use?
What happens when adversaries run their own orchestration layers against ours?
These are not procurement questions. They are force design questions, and they cut to the core of how militaries will be structured in the next decade.
Flow 3: The Strategic Question
With agentic AI, defence crosses from “faster staff work” to re-shaping command, coalition, and organizational design.
Previously, the challenge was cultural adoption.
Now, the challenge is structural re-design:
What is the commander’s role in an orchestrated system?
How do alliances certify AI-mediated interoperability?
How do states secure orchestration logic from adversarial interference?
These are governance and authority questions, they determine not just how AI is used, but who leads and who is accountable in a machine-mediated battlespace.
Flow 4: Doctrinal Shift
Traditional doctrine assumes humans own the loop: Observe → Orient → Decide → Act (OODA). With agentic AI, portions of Orient and even Decide are distributed into machine workflows. That is a doctrinal break. It forces militaries to decide how much agency can be delegated to orchestration layers and under what conditions.
Flow 5: Command and Control (C2)
Future C2 isn’t just humans issuing orders through communications systems.
Agentic AI layers act as micro-coordinators: synchronizing logistics, intelligence feeds, and cyber defences in real time.
Human commanders shift focus to intent and escalation points, while orchestration logic manages the flow beneath.
This reframes command authority itself. Instead of commanding tasks, commanders will increasingly command flows, trusting orchestration layers to execute while retaining judgment on what escalates.
Flow 6: Coalition & Policy
Agentic AI also changes the coalition equation.
This shifts interoperability from a data problem into an orchestration problem.
Policies will need to catch up with these new coordination logics: provenance, responsibility, auditability.
Alliances will need new standards not just for data sharing, but for certifying orchestration layers across nations. Without this, coalition partners risk running parallel but incompatible logics of coordination.
Conclusion – for now
The evolution of LLMs with layered capabilities is forcing defence to confront AI not at the edges, but at the core, in doctrine, in command, in force design.
“how do we govern machine-mediated orchestration at the core of defence?”
