Context: Dealing with professional complexity has been of scientific interest in the research field of vocational education and training for decades. So far, there is a lack of empirical evidence regarding how professional complexity finds its way into learning processes in VET. A common option is to model complexity through authentic simulations and/or problemsolving tasks. This study does both: Complex problem-solving tasks are integrated into an authentic office simulation and are expanded to include collaborative elements, using computer-based agents. Collaboration is used to improve learning, but it is also an authentic representation of current work processes, so we ask whether apprentices perform better in the individual or in the collaborative test setting. Methods: To compare skills used for complex problem-solving tasks within an authentic business simulation, the test settings are systematically varied between individual and collaborative test settings. The test environment is a technology-based assessment (TBA). Test tasks reflect a complex professional problem; they are presented in an office simulation, in which apprentices can collaborate with a computer-based agent. Multi-group confirmatory factor analysis is conducted to test for measurement invariance across test settings, and the two-dimensional Rasch-model incorporating between-item multidimensionality with correlated dimensions is used for ability estimation. We also conduct ANOVA tests to determine if there is a statistically significant difference regarding the problem-solving ability between the individual and collaborative test settings. Findings: The study provides various findings: First, collaboration helps apprentices to deal with complexity. Second, to solve complex problem-solving tasks, two bundles of skills are activated: Cognitive and social skills. The two-dimensional construct of skills with correlated dimensions showed better fit than the unidimensional construct. The scalar measurement invariance was established after excluding three items. ANOVA tests confirmed that the collaborative setting enhances the problem-solving ability of learners significantly regarding both, cognitive skills and social skills, with cognitive skills being fostered more. Conclusion: The findings suggest the validation of the two-dimensional construct with cognitive and social skills in economic domains. The results show the effectiveness of collaboration with a computer-based agent. In the practice of vocational training and education, teachers can use digitalized collaborative approaches to enhance learning.