Social robots and AI are increasingly valuable assets in various workplace settings, including healthcare, the service industry, and decision support systems. Understanding how trust differs between non-human and human teammates is crucial for creating effective and cohesive work environments. A key challenge is trust miscalibration, which involves the overreliance (misuse) on less capable agents and the underreliance (disuse) on more capable ones. Notably, teams with multiple non-human agents often exhibit system-wide trust, where all agents are perceived as having equal capabilities, increasing the risk of miscalibrated trust. It is essential to evaluate team members based on their specific abilities, known as component-specific trust, regardless of the task or agent involved. The presented research aims to validate new metrics for trust in human-agent interactions in both online and lab environments by aligning perceptions with actual capabilities. In a series of studies, participants collaborate with one, two, or three teammates (human, robot, or AI), with varying capabilities, to identify target symbols among similar distractor symbols in a visual search task. We use behavioral measures (e.g., interventions to assist teammates) and physiological indicators (e.g., eye-tracking or mouse movements to monitor teammates) to establish more objective trust metrics. Our findings suggest that this paradigm effectively distinguishes trusting behaviors in both monitoring actions and interventions, serving as an initial step toward exploring trust in human-agent teams more effectively.