Non-obstructive azoospermia (NOA) affects 5% of infertile couples, and current methods for isolating sperm for intracytoplasmic sperm injection (ICSI) are inefficient and prone to human error. This study investigates whether an AI-based image detection model, visualized through a microscope camera, can speed up and improve the accuracy of sperm identification in testicular tissue samples, thereby potentially enhancing ICSI success rates compared to traditional techniques. We conducted a side-by-side comparison of embryologists assisted by a trained object detection AI model versus unassisted embryologists (N=16). Key metrics recorded included the time taken and the number of ICSI-suitable sperm identified during each treatment, as well as treatment outcomes. Most samples were fresh microTESE, except one which was a frozen microTESE sample. Sperm searching was carried out using a conventional ICSI micromanipulator microscope, with and without AI assistance. Results were statistically analyzed using the Mann-Whitney U-test, with a p-value of <0.05 indicating significance. The AI-assisted embryologist demonstrated a reduction in search time per dish and identified more ICSI-suitable sperm than the unassisted embryologist. Specifically, the AI-assisted embryologists processed samples 33.1% faster and enabled more dishes to be searched, resulting in a 57.8% reduction in time per sperm found. The time taken per dish was significantly lower for the AI-assisted embryologist compared to the unassisted one (20.5±11.5 minutes vs 30.8±11.5 minutes, P=0.017). AI-powered image analysis has the potential to enhance laboratory workflows by reducing the time required for identifying and isolating sperm from surgical samples and mitigating fatigue from extended sperm searching. This increased efficiency may allow embryologists to process larger sample volumes and more dishes within the same timeframe, thereby increasing the likelihood of finding suitable sperm for ICSI.