The substrate layer below AI-augmented CAE

The visible AI interface is often the least interesting part of AI-augmented CAE. The harder work sits underneath: artifact indexing, lineage, permissions, retrieval, evaluation loops, and the boring reliability needed for engineers to trust a recommendation.

A simulation organization does not have one clean corpus. It has solver files, meshes, scripts, requirements notes, plots, spreadsheet summaries, issue threads, design-review decisions, failed runs, and local naming conventions. The substrate has to map that reality before the assistant can be useful.

Lineage is the difference between a helpful answer and an answer that cannot enter the engineering record. The system should preserve where a statement came from, which run it references, what changed between runs, and what evidence a reviewer can inspect later.

Permissions are not a cosmetic layer. Different programs, teams, suppliers, and review contexts may have different access boundaries. A CAE assistant that ignores those boundaries creates risk even when its language sounds technically fluent.

Evaluation also has to live below the interface. Teams need ways to test whether retrieval is complete, whether summaries remain faithful, whether recommendations cite support, and whether known failure cases are caught before engineers rely on them.

Barg Labs is building Offa for AI-augmented simulation engineering.

The product lesson is simple: the chat box is only the last mile. The durable system is the substrate that makes simulation artifacts addressable, reviewable, and governed enough for repeated engineering work.