Addy Osmani coins the term “ambient anxiety tax” for the part of running parallel AI coding agents that is not active cognition but background vigilance: the part of a person’s mind that cannot fully relax because it knows something might be quietly going sideways in a thread that has not been checked in a while. He describes it as drawing on the same finite reservoir as a person’s other cognitive work, even though it never shows up as an item on a task list of its own.
Usage
Osmani distinguishes the ambient anxiety tax from active cognition, introducing it as the background-vigilance component of running several agents in parallel that is easy to underestimate. Elsewhere in the same post he separately names three places where the work of running parallel agents actually lives — context switching (the recovery cost of reloading a mental model each time attention moves between agents), a continuous stream of judgment calls that interrupt rather than queue, and a trust-calibration overhead that degrades whenever a thread goes unwatched — without listing the ambient anxiety tax as a fourth item alongside them. He connects the ambient anxiety tax to what he elsewhere calls “comprehension debt” (Cognitive Debt): in a parallel session, that debt compounds across threads simultaneously, and a person cannot always tell which thread is running up the biggest tab.
Related Terms
Cognitive Debt, Orchestration Tax, Cognitive Surrender, Parallel Agent Limit