Implementation intentions (strategic if-then plans) have been shown to support behaviour change. This may be achieved by mentally forming stimulus-response associations, thereby promoting habit formation. Does this deliberate attempt to install ‘strategic automaticity’ only offer advantages, or does it also come at the cost of the reduced flexibility that characterizes learnt habits? To investigate this, we tested healthy, young participants on a computerized instrumental learning task. Critically, we introduced implementation intentions (e.g., if I see stimulus X, then I will respond) versus goal intentions (e.g., for outcome Z, I will respond) during instrumental acquisition, and subsequently assessed behavioural flexibility in an outcome-revaluation test. In Experiment 1, we conducted a between-subjects manipulation of strategic planning, and in Experiment 2 a within-subject manipulation. We hypothesized that implementation intentions would lead to strong stimulus-response associations and consequently impair performance when the signalled outcome value changed and therefore required a different response, while benefitting performance when the outcome value (and required response) remained the same. We found that implementation intentions supported instrumental learning more than goal intentions, but impaired test performance overall (most robustly in Experiment 2), irrespective of whether the signalled outcome value had changed. We argue that this general detrimental effect of implementation intentions on test performance is likely a consequence of its negative effect on stimulus-outcome learning. Our findings indicate that implementation intentions may support efficient and fast behavioural execution, but warrant caution in applying these to situations where the agent does not already possess perfect knowledge of behavioural contingencies.