Breaking Expert Work Into Repeatable Steps
Complex expertise becomes easier to execute consistently when it is divided into small decisions that people, models and tools can handle separately.
Developed from a conversation between Pete Winn, Paul Itoi and Andy David
Paul calls this approach factored cognition. The aim is to identify the smallest decisions inside an expert task, then turn those decisions into explicit workflow steps. Once the reasoning is visible, each step can be assigned to the person, model or tool best suited to perform it, rather than asking one worker or agent to manage the entire problem.
Stackwork first applied the principle to data entry from nutrition labels. Asking one person to transcribe a whole label meant tracking dozens of fields at once. When the team sliced each label into individual rows and gave each row to a different person, accuracy rose sharply. A worker only had to capture one bounded item, such as protein or fat, instead of maintaining attention across the full label.
The same method reduced a specialist task in salmon farming to a short decision path. Graduate students had been watching fish videos to identify several kinds of mites, but their judgement could be expressed through four or five observable checks, including the number of legs and the presence of a dark spot on the abdomen. Vision models could perform about half of those checks, while a workflow routed the remaining steps to other tools or people and fell back to a human when needed.