Since Euclid, the chain of logical deduction—from axiom to theorem to consequence—has been the way that human knowledge advances with certainty. Engineering inherited this tradition and industrialized it, making every bridge, airplane, and semiconductor the physical expression of a mathematical demonstration that it should work. But generative AI operates on a fundamentally different principle. Its outputs emerge from statistical patterns rather than logical derivations. It predicts, rather than proves. In many domains, that distinction hardly matters. But in others, small errors cannot be allowed to compound into failures. "The domains that most need AI's help—complex engineering workflows where human time is the binding constraint—are precisely the ones that can least tolerate its errors," says Jake Taylor, CEO of Axiomatic AI.
With deep scientific roots in the physics and engineering laboratories of MIT, ICFO, and the University of Toronto, Axiomatic AI is building the trust infrastructure needed to transform the use of AI for science and engineering. Its core technology harnesses frontier AI models with a structured knowledge base and flexible verification layer, enabling AI that does not merely reason about the physical world, but demonstrates that its reasoning holds. Axiomatic AI’s platform helps engineers accelerate their work and explore possibilities they couldn’t reach unassisted, without requiring them to trust AI outputs on faith. In practice, Axiomatic AI delivers agents and tools that can read the scientific literature, orchestrate simulations, optimize designs, and validate results against experimental data—working with human engineers to define the next generation of workflows across semiconductors, photonics, automotive, aerospace, and beyond. “The scientific method is our most trusted engine for discovery and validation," says Taylor. "We believe it deserves an AI counterpart grounded in the same principles of logic and evidence.”
Axiomatic’s co-founders are physicists and engineers who have spent their careers building complex systems, and advancing scientific knowledge through publication and teaching. They are a MacArthur grant winner, a White House policy veteran, and academic leaders across foundational AI, photonics, and semiconductor research. With the arrival of generative AI, they found themselves confronting a shared frustration: the new systems could describe their fields with impressive fluency, but couldn’t actually do new science, and struggled with the exactitude of engineering. Too often, wrong outputs cascaded, breaking the chain of understanding. Complex technical questions were answered confidently—and incorrectly. In the face of this persistent uncertainty, AI-assisted engineering seemed impossible. Even a small probability of error at each step makes the whole unworkable.
The co-founders coalesced around a founding insight: AI's tendency to hallucinate is a structural feature of probabilistic systems that can be engineered around. "If you can identify the errors and correct the errors, the better you can do that, the more depth your program can have, and the more sophisticated the system can become," explains co-founder Dirk Englund. They saw the opportunity to build AI that reads, reasons, links domains, and delivers interpretable designs. If errors can be corrected and outputs verified at each step, the application space of frontier AI expands to include complex systems. Axiomatic AI’s growing platform is guided by deep domain expertise and the accumulated weight of scientific knowledge, and functions with robust verification and validation capabilities. It fundamentally addresses the challenges of coordination, integration, and verification that have limited the practical application of AI in complex hardware development.
Axiomatic AI's first commercial applications are in photonics and semiconductor engineering, where design cycles are long, simulation tools are sophisticated but fragmented, and the cost of fabrication errors is high. Its AI agents plug into the existing workflows of foundries and design houses, helping engineers explore larger design spaces, validate results faster, and close the loop between simulation and physical measurement. But the platform's architecture is deliberately domain-general, built to extend across any design discipline dependent on the rules of physics. "If we are to make the most of today’s transformational technologies, there is far more engineering to be done than engineers can deliver," says Taylor. "Axiomatic AI’s platform unlocks work that was previously impossible."
Embedded in the company’s name is an ambition: to build AI whose outputs rest, like the best of engineering and science, on verifiable principles—axioms—rather than probabilistic guesswork.
