Where the term comes from in this material ## Reality gap in this channel's material The term appears in Dan Shipper's Dreamforce 2025 interview with Silvio Savarese, Chief AI Scientist at Salesforce [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@220]. In that conversation, Savarese explains how Salesforce stress-tests agents using simulated organizations with synthetic data — and he is explicit that closing the reality gap between those simulations and real-world deployment remains open work [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@676]. In other words, agents that perform well on synthetic, simulated benchmarks may still stumble when they face real enterprise data, real users, and real business conditions — and nobody in the material claims the problem is solved. [moment: 676] How to use the term ## Using the term Use reality gap when discussing the risk that evaluation in simulation overstates real-world performance. It sits alongside the interview's broader diagnosis of why enterprise agents stumble — siloed data and jagged intelligence [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@431] [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@485] — but it is specifically about the simulation-to-production divide. - The reality gap is the distance between how agents perform on simulated organizations with synthetic data and how they perform in real deployments. - Savarese names it as still open work — not a solved problem [^50a3b3fa-9ad2-59a3-9c76-ddfe74635066@676]. - It matters because enterprises stress-test agents in simulation before customers ever touch them, so a large gap means tests do not predict production behavior.