AI Engineering Models Produce Counterintuitive Designs Human Engineers Would Reject
"We have engineers that are using these workflows, using AI to explore and asking the AI to explore these hundred thousands of configurations overnight. And when they come the morning after, they're looking at the results, and then they're getting back to us and telling us, hey, did it— Very impressive. The AI model came up with a design that I would have never thought would be good. If you had shown me this design like this, I would have said, hey, scrap this, this is not gonna work. But actually, those designs are better than anything we could come up with."
About this episode
Nathan Labenz interviews Thomas von Schaumer, co-founder and US managing director of Neural Concept, a Swiss company deploying specialist AI models to accelerate automotive and aerospace engineering. Von Schaumer reveals that Jaguar Land Rover achieved a 30x speedup in aerodynamic design iteration, evaluating 1,500 designs daily versus 50 with traditional physics solvers, while battery suppliers compressed development cycles by 80% with superior performance outcomes. The conversation illuminates a profound competitive crisis: Chinese automakers complete new car development in 18-24 months compared to 40-60 months for Western manufacturers, creating existential pressure on legacy companies to adopt AI-native workflows. Von Schaumer explains Neural Concept's approach of training company-specific models on simulation and test data, enabling engineers to explore design spaces orders of magnitude larger than human intuition allows. Notably, these AI systems produce Move 37-style breakthroughs where counterintuitive designs outperform anything human engineers would have tried. The discussion extends to Formula 1 racing, where compute limits create a surprising governance model that handicaps winning teams, and where Neural Concept's models help engineers token-max overnight to explore thousands of aerodynamic configurations between races. Von Schaumer projects a two-year roadmap where leading manufacturers will achieve 50-60% cycle time reductions by orchestrating AI across crash, aerodynamics, thermal, and manufacturing constraints, breaking organizational silos. The episode frames engineering as another domain following the now-familiar pattern of intuitive physics learned from data, agentic optimization workflows, and eventual foundation models, with physical product abundance as the logical endpoint once reasoning AI meets specialist validation models.
Key takeaways
- Jaguar Land Rover increased aerodynamic design iteration from 50 to 1,500 evaluations per day using Neural Concept's AI, a 30x speedup announced at NVIDIA GTC 2024.
- Chinese automotive manufacturers complete new car development in 18-24 months versus 40-60 months for Western companies, creating existential competitive pressure.
- Formula 1 explicitly caps compute hours for aerodynamic simulation and handicaps winning teams with reduced allocations to prevent arms races.
- Battery thermal management suppliers achieved 80% faster development cycles plus 20% better cooling and 15% weight reduction using AI workflows.
- AI models generate counterintuitive designs human engineers would reject that prove superior, forcing domain experts to rethink physics intuitions.
- Von Schaumer projects 50-60% development cycle reduction within two years as AI orchestrates across crash, aero, thermal, and manufacturing constraints.
- Neural Concept trains company-specific models on simulation and test data today but is researching foundation models for aerodynamics and other physics domains.