The standard MBSE workflow in hardware development works like this: requirements live in a document, a systems engineer translates them into a Simulink model, a test engineer translates the model into a test plan, and someone runs the plan on a physical prototype six to eighteen months after the requirements were written. If the model was wrong (and it usually is, in some non-obvious way), you discover it when the prototype fails. MathWorks is making the case that AI-driven digital twins change this by closing the feedback loop before first silicon or first board.
The specific shift: digital twins updated by AI can incorporate real sensor data, process variation data, and test results from prior builds to self-correct their simulation fidelity. Instead of a model that is physically plausible but empirically uncalibrated, you get a model that tracks what the real hardware actually does. For embedded teams working on signal chains (radar, audio, motor control), the gap between what the model predicts and what the hardware does is where most re-spins live. Closing that gap with AI-updated models rather than waiting for hardware is a validation shift, not just a speedup.
The constraint being removed is the manual re-calibration step. Today, when a physical prototype diverges from a simulation, a systems engineer manually adjusts model parameters to match. That takes time, requires domain expertise, and can introduce new divergences. AI models trained on test data can do this automatically and continuously, keeping simulation and physical behavior aligned across the development cycle.
MathWorks holds dominant market share in MBSE tooling for automotive, aerospace, and industrial hardware. When they describe a shift in how MBSE workflows are being used, it is worth taking seriously. Teams that are still running pure simulation models without feedback from physical test data are leaving a systematic early-detection layer on the table. The cost of the gap shows up on your re-spin budget.