The article introduces GTA, a scalable framework for generating realistic tasks for web agents, addressing the limitations of existing benchmarks. This innovation aims to enhance the training and evaluation of web assistants by providing detailed, executable task trajectories across various domains.