In this channel's material, synthetic data appears in the context of testing enterprise agents before they reach customers. In the Salesforce interview, Silvio Savarese explains that his team builds simulated organizations populated with synthetic data in order to stress-test agents — evaluating how agents behave in realistic business scenarios before deployment, while acknowledging that closing the gap between simulation and reality remains open work [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@676]. How synthetic data is used ## Stress-testing agents with simulated organizations Savarese decomposes an agent into memory, a brain (reasoning engine), actuators (API and function calls), and an interface, and argues an agent without connected data and context is useless [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@220]. Synthetic data is one of the tools used to explore that behavior safely: - Simulated organizations provide a controlled environment where agents can be exercised against realistic enterprise conditions. - Stress-testing with synthetic data lets the team understand agent failures — including jagged intelligence, where models prove theorems yet fail at common-sense business tasks — without risking real customer systems [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@485]. - The reality gap between simulated and real deployments is, in Savarese's telling, still open work [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@676]. - Synthetic data, in this material, is generated data used inside simulated organizations to stress-test agents. - It connects to Savarese's broader point that enterprise AI struggles with siloed, poorly connected data [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@342]. - Simulation-based testing is presented as a work in progress: the gap to real-world performance is not yet closed [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@676].