Warmer mothers tend to produce smaller offspring across a broad array of organisms. The reasons for this pattern are unclear and most hypotheses have focused on the early stages of offspring development. Here, I use a marine bryozoan, Bugula neritina, which is a model system for understanding marine life histories. I report on a massive, ongoing field experiment in which I spawn adult colonies in the lab, measure the size of their larvae, and deploy around two hundred settled offspring into the field every six weeks. I use the local environmental temperature to measure how the relationship between initial offspring size and subsequent adult performance changes with temperature. After more than four years and the deployment of over 6000 individuals, I found that mothers in higher temperatures favour smaller offspring, while mothers in cooler temperatures favour larger offspring. This suggests that mothers alter the size of their offspring to maximise their own fitness. This can, however, sometimes be detrimental to the fitness of the offspring themselves. As a consequence of this long running field experiment, I was also able to estimate the temporal autocorrelation in selection on offspring size (how predictable selection on offspring size is over time). This predictability of selection over time is rarely estimated formally. Selection in one cohort is significantly, positively correlated with selection in the next cohort – in other words, selection changes but is predictable for at least the length of one generation. Surprisingly, I also found a significant negative correlation between selection in one cohort and selection two cohorts from now – so whatever is a ‘good’ phenotype now, is a ‘bad’ phenotype for their grandchildren.