AESCAPE 2026

The Evolution of Engineering Software

1.0: The Rule-Based Era (Manual Logic) Engineering Software 1.0 relied on explicit, rule-based programming. Engineers were required to manually define algorithms and logic to solve physical problems, resulting in rigid workflows heavily dependent on human intervention.

2.0: The Data-Driven Era (Neural Networks) Engineering Software 2.0 introduced machine learning to replace costly numerical simulations with rapid neural network predictions. While efficient, its effectiveness is constrained by the need for large, high-quality datasets that are expensive to generate, which highlights a critical need for automated data generation and calibration.

3.0: The Agentic Era (Mechanistic Language Models) We term this new paradigm Engineering Software 3.0. By leveraging foundational Large Language Models (LLMs) within an agentic framework, we dissolve the boundary between specification and implementation.

At the core of this shift is the Mechanistic Language Model (MLM). Fine-tuned with domain-specific mechanistic knowledge, MLMs establish an autonomous workflow that streamlines theory development, matrix algebra, and scientific code generation. Consequently, the MLM can autonomously generate geometry, mesh, and versatile code tailored to diverse engineering problems.

Why It Matters: This innovation reduces the toil of traditional numerical analysis: automating cumbersome tasks like CAD-to-mesh and post-processing, while enabling the solution of multiphysics problems previously deemed unsolvable.

Organizing Committee

Chair: Wing Kam Liu (Northwestern University, HIDENN-AI, LLC)

Co-Chairs:

• Sourav Saha (Virginia Tech)

• Miguel Bessa (Brown University)

• Thomas J.R. Hughes (The University of Texas at Austin)

• Michael Hillman (Karagozian and Case, Inc.)