From our latest Radical Talks episode with P-1 AI Co-founder and CEO Paul Eremenko
The AI revolution is changing nearly every industry, from software to content creation, yet one crucial area remains largely untouched: the design of physical things. While AI is great at manipulating digital bits, the physical world — from skyscrapers to spacecraft — still relies on traditional engineering approaches that haven’t fundamentally changed in decades.
This gap represents a huge, untapped opportunity. Physical systems could benefit enormously from AI-assisted design, but it’s a tough problem to solve. Current AI models struggle with quantitative and spatial reasoning, and training data is scarce. P-1 AI, a company co-founded by former Airbus CTO Paul Eremenko, is working to bridge this gap by developing an agent capable of mastering the physics and quantitative reasoning needed for physical design.
Why Is AI Not Used in Physical Design?
Despite decades of talk about AI transforming engineering, progress has been slow. While AI can write code and solve complex digital problems, it can’t yet design a new airplane wing or optimize a cooling system. This is due to two main challenges:
Limited Data: Unlike software, where millions of code repositories provide endless training data, the physical world has far fewer examples. Eremenko points out that only on the order of a thousand airplanes have ever been designed, a tiny number compared to the millions of examples needed to train an AI system.
Complex Reasoning: Physical design requires capabilities that current AI systems lack, such as quantitative reasoning, spatial understanding, and a deep intuition for physics, as well as the ability to use complex engineering tools effectively. These skills are very different from the next token prediction tasks at which frontier models currently excel.
The Breakthrough: Synthetic Data
To overcome the data problem, P-1 AI is using a clever solution: synthetic data generation. Instead of waiting for real-world designs, they use advanced modeling tools to computationally create millions of engineering designs validated with a suite of multi-physics simulation tools.
The key is a strategic approach to sampling these design spaces. They sample heavily around proven designs while also exploring the boundaries where new innovations can emerge. By combining this synthetic data with an ensemble of models, they can perform complex reasoning, ensuring their designs are both physically and spatially feasible.
Introducing Archie, the AI Engineer
The path forward requires specialized AI systems built for engineering. P-1 AI’s agent, named Archie, is able to utilize existing industry software while leveraging a multi-model architecture to handle specific engineering tasks.
In this sense, Archie functions as a junior engineer, performing core tasks like identifying design needs, generating solutions, and selecting analysis tools. Importantly, Archie doesn’t replace existing engineering software; instead, it automates the complex reasoning that connects them and works with these existing tools. As Eremenko notes, “The vast majority of things that design engineers do are reducible to a few primitive operations.” This modular approach allows specialized AI models to excel at specific tasks while coordinating through a unified interface.
Real-World Applications
P-1 AI is focused on commercial applications that are both technically feasible and have strong market demand. A great example is data center cooling systems. These systems involve multiple areas of physics but are less complex than aerospace designs. The timing is perfect, as the growing demand for data centers is creating an acute need for more efficient designs.
The broader opportunity for AI-assisted design spans across industrial equipment, mobility, and, eventually, aerospace and defense. The manufacturing industry is under pressure to adopt these tools to stay competitive, meet sustainability goals, and keep up with rapidly evolving technology.
The Future: General Engineering Intelligence
P-1 AI’s ultimate goal is to create engineering artificial general intelligence, an AI capable of designing across multiple domains and physics regimes, much like a human engineer can transfer knowledge from one project to another.
While formidable technical challenges remain, the potential is enormous. This technology could accelerate human civilization’s ability to design physical systems by orders of magnitude, transforming the world around us.
Building the Future We Want
Engineering AI systems like P-1 AI promise to make teams more productive and agile in a rapidly changing world. More importantly, they may help us design entirely new kinds of physical systems that we don’t yet understand how to create.
“The future is not a thing that happens. The future is a thing that we build together,” Eremenko concludes. This philosophy drives research toward artificial general engineering intelligence as a tool for expanding human capabilities.
As AI systems become increasingly capable of reasoning about the physical world, the engineering profession may undergo its most significant transformation since computer-aided design. The companies and researchers bridging artificial intelligence and physical design are laying the foundation for humanity’s next chapter of technological capability.
This post is based on insights from Radical Talks, a podcast from Radical Ventures exploring innovation at the forefront of AI. For more cutting-edge AI discussions, watch full episodes on Apple Podcasts, Spotify, or YouTube.