Antibiotic treatment failure, associated with non-genetic mechanisms such as tolerance and persistence, remains a global health challenge. β-Lactams, the most widely prescribed antibiotic class, are particularly affected by tolerance and persistence, but the mechanisms are not fully understood. Here, we investigate persistence during resuscitation from starvation in E. coli using a high-throughput single-cell microfluidic platform combining time-lapse imaging and machine-learning-based analysis. We identify a surprisingly prevalent survival strategy: a considerable fraction of resuscitating cells transiently slow growth during antibiotic treatment, facilitating their survival. This phenotype is virtually absent in unstressed, exponentially growing cells, suggesting that it is programmed by a “stress memory” from prior starvation. Its frequency scales with starvation duration, and effective killing of this class is highly dependent on antibiotic concentration and exposure time. By mimicking antibiotic pharmacokinetics, we show that clinically relevant treatment profiles of common β-lactam antibiotics fail to eliminate this prevalent and actively growing survivor class. These results demonstrate that these transient growth-slowdown persisters, not classical dormant persisters, are the primary drivers of rapid population regrowth following β-lactam treatment in vitro.