Cogs of War Agentic AI and the Pentagon's Integration Challenge
Feb 24, 2026
Aaron Brown, former CIA officer and CEO of Lumbra AI, builds orchestration for agentic workflows. Jags Kandasamy, CEO of Latent AI, compresses models for edge and tactical use. Ben Van Roo, CEO of Legion Intelligence, focuses on hardened deployments. They discuss running agentic AI on military networks, compressing models for the edge, orchestrating across stovepipes, and reviving legacy systems for real operations.
AI Snips
Chapters
Transcript
Episode notes
Agentic AI Is Automation With Decision Power
- Agentic AI adds decision-making degrees of freedom beyond deterministic automation.
- Jags Kandasamy explained agents choose next steps (like taking different elevator floors) rather than following fixed if-then flows, enabling workflow intelligence.
Decide Human Involvement By Response Time
- Design agent deployment by ruling where humans sit: in the loop, on the loop, or out of the loop.
- Jags advised using doctrine and scenario-specific response-time thresholds to decide when agents should act autonomously.
Agents Can Be More Auditable Than Monolithic LLMs
- Agentic systems can be more auditable than big opaque LLMs when designed as workflow-driven agents.
- Aaron Brown said agents can map steps to doctrine and provide receipts back to commanders for explainability.