What's under the hood? :A transparent V&V led framework for AI surrogates in engineering simulation

AI-based surrogate models are becoming a practical way to speed up engineering simulation, but only if they can be trusted like established analysis tools. Using a recent nearshore wave study, this contribution shows how teams can build and deploy AI surrogates that behave predictably, can be checked against known results, and remain safe to use when conditions change. The focus is on methods that fit naturally into day-to-day engineering workflows, not specialist AI techniques. To support trustworthy use, the workflow keeps things clear and testable. Before building a surrogate, engineers define intended use, operating envelope, and out-of-scope conditions. These expectations drive acceptance checks: comparisons with a trusted solver, targeted limit and stress tests to expose weak spots, and automated regression tests to catch unintended changes when the model is updated. Put simply, verification asks "Did we build the model right?" while validation asks "Did we build the right model?" and the checks are organised to answer both. All training data is logged so results can be traced to their sources and assumptions reviewed when requirements shift. Together, these steps form a clear, repeatable way of working that teams can adopt and build on as part of their own good practice. In the wave modelling example, the surrogate is exercised across calm, moderate, and challenging sea states, with tests placed near boundaries where the physics becomes more sensitive. This shows engineers how the model behaves under pressure, not just in ideal conditions. We also show how to probe for data shift by withholding specific scenarios and making controlled variations to key inputs so the surrogate's limitations are visible, quantified, and communicated. This pattern: define, test coverage, compare to references, track changes, applies to any simulation task where speed is useful but decision grade confidence is essential. The result is a practical, repeatable checklist of actionable steps engineers can apply across domains to gain the speed benefits of AI while maintaining the level of trust and assurance expected in engineering simulation.