The proliferation of enthusiasm into the near and far benefits of working memory (WM) training in recent years has largely been driven by the assumption that WM capacity represents a domain-general limited control mechanism. Proponents of such interventions argue that by continuously taxing such basic processes through adaptive procedures tailored to an individual’s current ability level, there is an increased likelihood of transfer to disparate cognitive domains. However, the field has recently been subject to various critiques. The present study addresses a number of design limitations and answers numerous calls to take a theoretical approach to the assumptions underlying the recent wave of cognitive training research. Participants completed 10 sessions of online WM training that differed with regard to an adaptively tailored (experimental) or randomly presented (control) difficulty level. No differences in transfer to measures of WM, processing speed or fluid intelligence were observed between the two groups. Taking a multicomponent approach of WM (Logie, 2011), the change in the ability to recall both phonologically and visually similar sequences of letters was explored. No reliable evidence of change in the use of domain-specific components was observed. There was a strong involvement of phonological storage to support recall. The need for the field to define appropriate control protocols as well as empirically establish the efficacy of the features and mechanisms currently endorsed in the literature is discussed.