Building the Physical AI Stack | Travis Kalanick on TBPN

What It Covers

Travis Kalanick discusses his new company Atoms, which raised $1.7 billion to build industrial AI systems that automate physical industries including mining, food production, and transport. He covers go-to-market strategy for enterprise hardware, executive hiring frameworks, autonomous mining operations, and the economic case for physical automation over software-only businesses.

Key Questions Answered

• •Industrial AI go-to-market: Enterprise physical automation requires in-person proof before scaling. Kalanick visits active mine sites — including a Vale iron ore operation in the Brazilian Amazon and a phosphate mine on the Iraq-Saudi border — to demonstrate that Pronto's autonomous haulage systems exceed human productivity benchmarks before customers commit to full fleet deployment.
• •Autonomous mining productivity gains: Retrofitting existing mining vehicles with sensors and compute to enable autonomous haulage can yield 30–40% productivity increases per mine. Gains come from two sources: machines operating more efficiently per hour, and eliminating shift callouts and safety downtime. This math applies across gold, lithium, iron ore, and quarry operations.
• •Retrofitting vs. native autonomy: Most mining equipment is not drive-by-wire, meaning physical actuators must be installed to convert mechanical and hydraulic steering systems into software-controllable ones. This commissioning process is the primary scaling bottleneck — not customer demand — making installation speed and change management the core operational challenge for industrial AI deployment.
• •Executive hiring framework: Kalanick prioritizes problem-solving ability over organizational management skill when hiring executives. His reasoning: a strong organizer who cannot solve problems executes bad decisions efficiently. His interview process simulates actual working conditions so that day one functions like week two, reducing first-90-days failure risk and validating problem-solving capacity before hire.
• •Business model for outcome-based hardware: Kalanick structures industrial AI pricing like enterprise software — a baseline subscription with outcome-linked upside. He advises against asking customers for revenue percentages directly, instead setting a fixed price with performance bonuses. The principle: always create more value than you capture, and let differentiation determine how much additional margin you can negotiate.