Physical AI that Moves the World — Qasar Younis & Peter Ludwig, Applied Intuition

Applied Intuition focuses on deploying intelligence into physical machines, including cars, trucks, construction equipment, and defense technologies. Unlike large language models that operate on screens, this technology functions in safety-critical environments where errors are unacceptable. The company evolved from providing simulation and data infrastructure for robotaxi companies to becoming a comprehensive technology provider for the industrial sector. With over 80% of the workforce in engineering, the organization prioritizes recruiting talent with deep expertise in the intersection of hardware and software, maintaining a strategic focus on long-term industrial advancement rather than consumer-facing applications.

The technology portfolio consists of three primary pillars: simulation infrastructure, operating systems, and fundamental AI models. Simulation is critical for testing complex software in virtual environments before real-world deployment, while the operating system layer manages real-time control, memory, and highly reliable software updates. This OS architecture is designed to handle the latencies and safety requirements of embedded systems, enabling manufacturers to move beyond fragmented firmware. By providing a unified platform, the company enables industry-wide software updates, a capability that remains rare in traditional vehicle manufacturing and is essential for maintaining safety-critical systems.

The physical machine industry currently mirrors the fragmented state of the phone market prior to the emergence of Android and iOS. By consolidating the operating system, manufacturers can deploy modern AI applications across diverse hardware architectures. Internally, the adoption of AI coding tools has created a bimodal distribution of engineering productivity, where those who master these tools significantly outperform others. While embedded systems were once considered resistant to AI integration, modern models now provide immense value in writing complex code and configuring sensor suites, provided that human validation remains central to the safety-critical development process.

Verification and validation have shifted from binary, requirement-based testing to statistical models focused on reliability and mean time between failures. As autonomous systems become more complex, proving safety requires demonstrating multiple nines of reliability. While public perception is often influenced by high-profile accidents, these incidents are frequently compounded by organizational failures rather than purely technological ones. Statistical evidence confirms that autonomous systems are safer than human drivers, who are prone to fatigue and impairment. Long-term adoption depends on society accepting that while accidents may occur, the overall safety profile of autonomous systems is superior to human-operated transport.

Bridging the gap between simulation and reality requires iterative feedback loops where real-world data informs simulation parameters. World models are increasingly used to understand cause-effect relationships, such as how environmental conditions like rain affect vehicle control. However, relying solely on world models is insufficient for production; practical engineering requires a hybrid approach. On-machine compute constraints necessitate model distillation and extreme efficiency, as every millisecond of latency impacts vehicle performance. Transformers have become universal, but their application in embedded systems requires careful optimization to balance capability, power consumption, and the harsh physical conditions in which these machines operate.

Physical autonomy tasks, from driving to excavating, can be modeled as sequences of steps or trajectories, making them compatible with next-token prediction techniques. Building a successful deep-tech company requires a focus on compounding technology where each advancement—whether in dev tooling, OS, or models—builds upon the last. Founders should avoid broad, shallow strategies and instead apply first-principle thinking to solve specific, constrained problems. Success in this space demands extreme curiosity about fundamental physics and low-level systems, as well as the ability to navigate the long, difficult path of commercializing technology in industries that require rigorous safety standards and long-term commitment.