agentsfactneutralThe field is increasingly turning to VLMs as judges of CUA trajectories for evaluation, data curation, and reinforcement learning at scaleComputer Vision02 Aug 2026http://arxiv.org/abs/2607.28609v1