1000 Designs a Day: Neural Concept's Thomas von Tschammer on AI-Native Engineering

What It Covers

Neural Concept cofounder Thomas von Tschammer explains how AI surrogate models replace physics-based simulation solvers in automotive engineering, enabling Jaguar Land Rover to evaluate 1,500 aerodynamic designs daily versus 50 previously, while an agentic engineering copilot automates CAD modifications and cross-disciplinary optimization across crash safety, thermal management, and aerodynamics.

Key Questions Answered

• •Simulation Speed Multiplier: AI surrogate models reduce physics simulation time from days to minutes, enabling manufacturers to evaluate thousands of design configurations daily instead of dozens annually. Jaguar Land Rover moved from 50 aerodynamic designs evaluated per day to 1,500 in production. Battery cold plate suppliers reduced development cycles by 80% while achieving 20% better cooling performance and 15% weight reduction simultaneously.
• •Per-Customer Model Training: Neural Concept trains domain-specific models on each company's proprietary simulation and physical test data rather than deploying generic foundation models. This approach captures company-specific engineering know-how, and models continuously improve as new data is added. Every simulation run feeds back into retraining, compounding accuracy gains across successive product development cycles and preserving institutional knowledge.
• •Agentic Workflow Architecture: The engineering copilot combines frontier LLMs with domain-specific physics models as callable tools, enabling automated CAD geometry modification, simulation submission, and results analysis. Year-one targets for automotive OEMs should focus on AI-led iteration within single disciplines like crash or aerodynamics, targeting 20-40% cycle reduction before expanding to cross-disciplinary orchestration in year two.
• •Cross-Disciplinary Orchestration as the Real Multiplier: Breaking silos between aerodynamics, crash safety, thermal management, and manufacturing constraints within a single agentic workflow produces compounding gains beyond single-discipline optimization. Neural Concept reports 50% development cycle reductions where multidisciplinary automated workflows are deployed. Chinese automakers currently complete new vehicle development in 18-24 months versus 48-60 months for Western OEMs, a gap AI orchestration directly targets.
• •Formula One as Engineering Stress Test: Formula One teams serve as validation environments for AI engineering workflows because cars are redesigned between every race weekend. Regulatory compute caps on CPU hours for aerodynamic simulation create a direct incentive to maximize design quality per simulation run, making F1 teams ideal customers for surrogate models. Practices proven in F1 environments transfer directly to traditional OEM development processes.
• •Manufacturing Constraints Must Enter the Design Loop Early: AI-generated designs must incorporate stamping tolerances, injection molding rules, and production cost constraints from the first iteration, not as post-design filters. Embedding manufacturing design rules directly into surrogate model training prevents generating physically optimal but unproducible geometries. Suppliers who use freed development time to optimize manufacturing processes can reduce part costs enough to win additional programs worth tens of millions of dollars.