What a DARPA fighter-pilot project taught us about Unless

Version 1.0 · Published 2026-07-20

Picture a test operator in a simulator, somewhere around 2001. They are wired up with sensors that read brain activity and stress levels, working through a military decision task while a computer watches their cognitive state in real time. When the pressure spikes, the computer steps in and gets more directive. When things calm down, it backs off and leaves the decision to the human.

This was DARPA’s Augmented Cognition program, started in 2001 by the research arm of the American military. Its goal was to improve how warfighters make decisions when they are overloaded and under stress, with fighter pilots as the vivid example. Just before we started UNLESS, we heard about this project, and it never let go of us.

A computer that reads the room

The core idea was almost rude in its simplicity: instead of training the human to cope with the machine, make the machine adapt to the human. The sensors estimated how much cognitive load the operator could handle at that moment. The system then adjusted how much support it gave, from a light touch to taking real work off the operator’s plate.

And it worked, at least in the lab. In tests and simulations, including a naval command-and-control task, decision-making improved when the machine adapted its behavior to the person’s state. The principle was sound. The reader in you may already sense the “but” coming.

The part that failed

The program struggled with the human-machine interface of its time. Reading someone’s state through early-2000s sensors was clumsy, and the interfaces that were supposed to adapt were rigid and slow. The idea was right, but the technology around it was not ready to carry it.

That gap sat there for about twenty-five years. Then machine learning, natural language processing, and modern AI arrived, and suddenly a system could understand a person’s situation without wiring them up to electrodes. Plain language turned out to be the sensor DARPA never had.

From cockpits to customer journeys

Most people are never in a cockpit, but everyone knows the feeling the program was built for. You are on a payroll platform three days before the run, or on a patient portal reading something that worries you, and the interface gives you the same static screens it gives everyone else. The software makes you do the adapting, at exactly the moment you have the least capacity for it.

That is the problem we took from the DARPA story into Unless. If software can assist people in a personalized way, all the way through their journey, based on their background, their state of mind, and the problem at hand, it can improve their understanding and help them through difficult decisions. So we built the Customer Agent to do exactly that: one agent that knows who it is talking to and what they are dealing with, across acquisition, retention, expansion, and support.

The mechanics differ from 2001, but the shape is the same. Where DARPA used physical sensors, Unless runs on three senses: Living Knowledge for what the agent knows, Living Memory for who it is talking to, and Living Context for where it is acting. Pressure still gets read, just from language and situation instead of electrodes.

The bet, twenty-five years later

The lesson we took from that program fits in one line: good software should adapt to the person, not the other way around. DARPA had the right idea and the wrong decade. We get to build it in the right one.

That is the bet behind Unless. Not that AI can answer questions, which by now is table stakes, but that it can finally close the loop the Augmented Cognition researchers opened: a machine that notices how you are doing and meets you there. Somewhere, hopefully, a former DARPA researcher is nodding.