Effectively harnessing computer vision techniques powered by supervised deep learning in precision agriculture requires extensive data gathering and meticulous ground truth annotation that involves human effort. The emerging vision-based foundational models in Artificial Intelligence (AI) are gaining notable importance. Recently, Meta AI Research introduced a general, promptable Segment Anything Model (SAM) that has undergone pre-training on an extensive dataset dedicated to segmentation (SA-1B). Without additional training, this model is known to demonstrate its ability to generalize effectively to novel objects. This research assesses the effectiveness of the Segment Anything Model (SAM) in precision agriculture, specifically in counting crop heads within cereal crops such as paddy, wheat, and maize.