Why do we divide (‘chunk’) long tasks into a series of shorter subtasks? A popular view is that limits in working memory (WM) prevent us from simultaneously maintaining all task relevant information in mind. We therefore chunk the task into smaller units so that we only maintain information in WM that is relevant to the current unit. In contrast to this view, we show that long tasks that are not constrained by WM limits are nonetheless chunked into smaller units. Participants executed long sequences of standalone but demanding trials that were not linked to any WM representation and whose execution was not constrained by how much information could be simultaneously held in WM. Using signs well-known to reflect beginning of new task units, we show that such trial sequences were not executed as a single task unit but were spontaneously chunked and executed as series smaller units. We also found that sequences made of easier trials were executed as longer task units and vice-versa, further suggesting that the length of task executed as one unit may be constrained by cognitive limits other than WM. Cognitive limits are typically seen to constrain how many things can be done simultaneously e.g., how many events can be maintained in WM or attended at the same time. We show a new aspect of these limits that constrains the length of behaviour that can be executed sequentially as a single task-unit.