Eli KaganMost AI interfaces are still built for conversation.
The real challenge is building for comprehension.
As language models continue to mature, the interface design challenge is no longer simply about enabling users to “chat with AI.” It’s about building systems where the agent demonstrates real contextual awareness, where it can interpret evolving operational states, synthesize noisy data, and present only the most relevant information, exactly when needed.
At USAvionix, we’re building Phalanx, a command layer for autonomous ISR missions.
Fleets of aircraft take off, coordinate, gather surveillance, and analyze findings on their own. The human operator supervises all of this through a single system that must balance comprehensibility, trust, and control at scale.
This is not a chatbot bolted onto a dashboard.
SARA (Strategic, Autonomous, Reconnaissance Agent), the embedded AI agent, is designed to act more like an embedded analyst, understanding the mission, the filtering assumptions, the sensor context, the historical intelligence, and the downstream implications of every decision.
The UI is built around interaction with judgment, not just access to capabilities.
The opportunity (and the complexity) lies in merging human cognitive models, expectations, workflows, intuitions, with machine inference models.
- How do you reduce a live, multi-aircraft autonomous op to a single sentence without omitting anything mission-critical?
- How do you surface emergent threats without creating alert fatigue?
- How do you allow for oversight without requiring micromanagement?
These are interface problems.
But they are also epistemological problems, about what the system knows, how confidently it knows it, and what it chooses to reveal. We are no longer designing software for people to operate machines.
We are designing systems where software and machines operate in concert, and people supervise the system.