Goal-directed behavior requires an adaptive balance between goal persistence and disengagement. The present research investigated how a reliable vs. unreliable co-actor affected participants’ goal persistence in a joint setup. In a novel paradigm, participant and a confederate moved the target in two steps from the bottom center to the top left or right corner of the screen. First, participants moved the target halfway towards either corner. Then, the confederate moved the target to its final position. In the first block, the confederate always followed the participant’s choice. In Experiment 1, the confederate followed the participant in 50% (unreliable follower) vs. 80% (reliable follower) of trials in the second block (between). In Experiment 2, the confederate followed the participant in 50% of trials in the second and in 80% of trials in the third block (or vice versa; within). In Experiment 3, the confederate changed from block 2 to block 3 in addition. As DV we used the proportion of goal repetitions in dependence of whether the confederate had followed the participant’s choice or not. In Experiments 1 and 3, results show increased goal persistence (goal repetitions) after the confederate did not follow the participant’s choice. Crucially, this effect was stronger when the confederate was a reliable (80%) rather than an unreliable (50%) follower. Note that these effects were absent in Experiment 2. We conclude that interdependent goals are only maintained when the co-actor is perceived as reliable.