Definition
Plain language
A moment in an agent's task where it has to choose between using a tool and continuing with GUI clicks.
As stated in the literature
In ToolCUA training, a step in a synthesized hybrid trajectory where a tool call was swapped out for GUI actions and the corresponding tool removed from the library, manufacturing an explicit mode-selection decision for single-turn RL.
Also called: critical switching steps
Why it matters: It manufactures the exact decisions an agent needs practice on — choosing between tools and direct interaction — instead of hoping they appear naturally in data.
For example, a trajectory that originally clicked through five GUI screens to change a setting is rewritten so that the agent must decide between calling a 'set_preference' tool or proceeding with clicks.
Heard on the show
“Every swap creates what the paper calls a "critical switching step" — a moment where the agent has to make an explicit choice between modes.”Episode 066 — Why Giving an AI Agent More Tools Can Make It Worse at Using a Computer