Task-Set Reconfiguration

In the AI Story

The mechanism is distinct from attention residue but complementary. Monsell's task-set reconfiguration describes the forward-looking cost of preparing for the new task; Leroy's attention residue describes the backward-looking cost of the old task's persistence. Together they constitute the full cognitive tax of switching: the executive system must both disengage from the previous configuration and engage with the new one, and neither operation is instantaneous or complete. The builder switching from evaluating AI-generated code to reviewing a design mockup must deactivate the analytical, detail-oriented, correctness-focused configuration appropriate for code review and activate the holistic, aesthetic, user-oriented configuration appropriate for design evaluation. The switch is not a toggle. It's a migration.

AI-augmented work increases both the frequency and the severity of reconfiguration demands. Frequency increases because rapid output production creates more switching events per day. Severity increases because the cognitive distance between tasks is often greater: evaluating backend code and frontend design mockup and audio processing pipeline and business strategy document within the same afternoon requires four dramatically different executive configurations. Each reconfiguration is expensive; each is incomplete when the next switch arrives; and the incomplete prior configurations contribute to the residue load that degrades the current evaluation. The builder is reconfiguring constantly and never fully configured for any single task.

The organizational implication is that not all context switches are equally expensive. Switching between similar tasks — two coding problems, two design reviews — requires smaller reconfigurations than switching between dissimilar tasks — from code to design to strategy to interpersonal conflict. Vector pods and other team structures that cluster similar evaluations can reduce total reconfiguration costs by allowing executive control to maintain a stable configuration across multiple similar tasks. The builder who evaluates three AI-generated code samples in sequence incurs three attention residue events but only one major task-set reconfiguration. The builder who evaluates code, then design, then strategy in sequence incurs three residue events and three reconfigurations, and the aggregate cognitive cost is substantially higher.

Origin

The concept originated in cognitive psychology's study of task-switching costs, particularly Stephen Monsell's 2003 review article synthesizing two decades of experimental findings. Monsell demonstrated that switch costs are not eliminable through practice — a finding that parallels Leroy's discovery that residue is not trainable. Both reflect fundamental features of cognitive architecture. The application to AI-augmented work recognizes that the diversity of outputs builders must evaluate — spanning code, design, strategy, and interpersonal domains — requires more extensive reconfigurations than the homogeneous task-switching that most experimental paradigms studied.

Key Ideas

Non-instantaneous switching. Even for simple tasks, executive control reconfiguration takes measurable time, and for complex knowledge work the interval extends to seconds or tens of seconds per switch.

Progressive displacement. The new task-set doesn't fully activate on the first trial after a switch; performance improves across several trials as the new configuration gradually displaces the old.

Diversity amplifies cost. Switching between cognitively distant tasks (code to design to strategy) requires more extensive reconfigurations than switching between similar tasks, multiplying the switching tax.

Clustering reduces costs. Batching similar evaluations allows executive control to maintain a stable configuration across multiple tasks, reducing total reconfiguration demands relative to alternating between diverse tasks.