Bottlenecks

The Abstraction and Concretization Skills You Need to Work AI-Natively

Coding agents make individual work faster, yet organizations do not get faster. In METR's experiment, developers believed they were 20 percent faster while measured time was 19 percent slower, and a 2026 survey found self-reported "speed" gains of 3x against "value" gains of only 1.4 to 2x. Two causes: people speed up work that is not the bottleneck, and the more context an AI holds, the more it drifts toward answers that fit that context and away from the abstract answer. This article explains how to split the abstraction-concretization loop between humans and AI, and how to train the skill of noticing what is not working and putting that unease into words, with primary sources throughout.