Google investigated AI agents which were capable of employing self-attention bottlenecks. Such agents are successful in tackling reasonable modifications to tasks and solving vision-based challenges. Google built AI based AttentionAgent that works by ignoring distractions and focusing on task-relevant factors. The agent divides the inputs into patches and decides a subset. These patches then help the AttentionAgent to act based on the changes to the input data and keep track of the evolving task-relevant factors.