Retrieval efficiency determines whether long-term memory supports or disrupts working memory

Goal-directed behavior depends on interactions between visual working memory (WM) and visual long-term memory (LTM), yet it remains unclear whether maintaining information in WM is compromised by concurrent retrieval from LTM or whether both processes can proceed in parallel with minimal interference. Influential accounts posit that WM is a mandatory bottleneck for LTM access, such that LTM retrieval constrains WM processing. In contrast, alternative views propose that LTM can guide behavior independent of WM. Across two experiments (N = 197), we tested whether and how LTM retrieval of learned spatial context shapes WM dynamics. We leveraged trial-level reaction times as a continuous index of LTM retrieval efficiency during contextual cueing of visual search under concurrent visuospatial WM load. Our findings show that repeated contexts enhanced WM performance relative to novel displays on fast-search trials but impaired WM performance on slow-search trials. Thus, the direction of the contextual effect on WM reversed as a function of trial-level LTM retrieval efficiency. Crucially, this reversal occurred within identical repeated displays, ruling out explanations based on stable stimulus properties and indicating that trial-by-trial variability in LTM retrieval efficiency determines whether contextual memory facilitates or interferes with concurrent WM maintenance. We propose that LTM guides attention with minimal additional processing when LTM retrieval is robust but engages additional control mechanisms when contextual signals are weak or ambiguous. Together, these findings argue against an obligatory WM bottleneck and instead support a flexible architecture in which LTM-WM coupling is dynamically modulated by LTM retrieval efficiency.