The Wrong Side of the Equation: Cybernetics for the Agentic Era

In 1973, the cybernetician Stafford Beer watched institutions buy computers and use them on the wrong side of the variety equation, pumping untold variety into systems designed to handle the output of a hundred quill pens.

We're doing it again. AI made writing code far faster, while reviewing still runs at human speed. In my longitudinal research, professional developers perceived themselves spending less time on nearly every part of the job except reviewing. They also described a new kind of work emerging with AI - supervisory engineering work - directing, evaluating and correcting code they didn't write, while remaining accountable for every line of it. With agents producing code faster every day, the job is becoming what automation researchers described more than forty years ago as an impossible task.

Cybernetics helps explain why: only variety can absorb variety. A human-speed regulator cannot keep pace with a machine-speed system. That leaves only two options - amplify the regulator, or reduce the variety it has to absorb - and it's worth asking which one we're actually choosing.

This talk makes the case for reimagining the SDLC as a control system to be engineered rather than a process to be followed, and explores how formal verification might become one of the instruments that keep it on course.