Advances in artificial intelligence (AI) have expanded firms’ ability to implement process controls beyond traditional output-based monitoring, yet their implications for labor productivity remain unclear. We examine how AI-enabled process control affects employee productivity using field data from a stevedoring firm operating at a major seaport. In our setting, AI-enabled cameras installed on gantry cranes generate both decision-facilitating signals that enhance visual guidance and monitoring signals that support behavioral oversight. Using a difference-in-differences research design, we find that AI camera adoption is associated with a decline in crane-operating productivity on average. However, decomposing AI signals reveals substantial heterogeneity. Decision-facilitating signals are associated with productivity gains when informational constraints are more severe, such as during complex cargo handling, night shifts, or among less-experienced operators, whereas monitoring signals are consistently associated with productivity reductions. Qualitative evidence from interviews corroborates these patterns, suggesting that AI monitoring induces cautious operating behavior, while decision-facilitating information improves task precision. Overall, our findings highlight the importance of distinguishing between information provision and behavioral surveillance when evaluating the productivity effects of workplace AI.