Caitlin Kalinowski on Building Hardware, the Robotics Frontier, and Supply-Chain Reindustrialisation

Caitlin Kalinowski on Building Hardware, the Robotics Frontier, and Supply-Chain Reindustrialisation

transcriptlennyhardwareroboticssupply-chainai

Caitlin Kalinowski on Building Hardware, the Robotics Frontier, and Supply-Chain Reindustrialisation

Source: Lenny’s Podcast Speaker: Caitlin Kalinowski Date: ~May 2026 Link: Episode

One of Silicon Valley’s most accomplished hardware leaders — original Unibody MacBook Pro, MacBook Air and Mac Pro lead at Apple; VR then AR hardware (Rift, Quest, Orion) at Meta; founder of OpenAI’s robotics/hardware effort — on why hardware and robotics are suddenly the frontier, what makes hardware uniquely hard, and the supply-chain reindustrialisation she sees as a national-security need.

Key ideas

  • The physical world is the next frontier. As what you can do “behind a keyboard” with AI saturates, labs, big tech, and startups are pivoting to robotics, manufacturing, and industrialisation. The VR/AR stack (SLAM, depth sensing, spatial perception) was the technological lineage that now feeds robotics, drones, and autonomy.
  • Hardware ≠ software: you only “compile” four or five times — ever. No over-the-air fixes once mass production starts, so you must front-load conservatism and reliability, and design for ±3σ part variance. Her design discipline: set goals/KPIs up front and barely change them; design the hardest/riskiest part first; over-iterate the parts the user touches most; and do everything you can now (in hardware you genuinely don’t have spare time). Use off-the-shelf parts in prototyping; go custom only where a KPI demands it.
  • Supply chain is the binding constraint — and a national-security one. Magnets → actuators → robots, each layer outsourced to Asia over 25 years. She calls for reindustrialisation (“drones over aircraft carriers”), warns of the “memory-price meteor” (AI/data-centre demand spiking DRAM; pre-buy to ride out shocks), and notes a single missing chip or RAM part forces a catastrophic redesign — the case for Tesla/Starlink-style verticalisation.
  • Humanoids are over-hyped; specialised robots win most tasks. A dedicated screw-driving robot beats a generalist humanoid; advanced manufacturing lines already run with ~10 people, not 200, so the need is more dedicated robots, not human-shaped ones. Safety (mass, compliance, signalling intent) gates humanoids near people; robots that “feel human” borrow Pixar/Disney social cues.
  • AI in hardware: helpful, not yet transformative. AI accelerates planning, Excel, databases, and PCB routing, but real parametric CAD needs world models that understand friction, contact, and weight — “I want Codex for hardware engineering.” The bottleneck is data: CAD is jealously-guarded IP, so hobbyists (who don’t mind sharing) may seed it.

Topics covered

  • Why VR didn’t take off, and how its stack (SLAM, depth) feeds AR and robotics
  • Hardware vs software: the “four or five compiles” model and design discipline
  • Robotics supply chain: magnets, actuators, reindustrialisation, the memory-price meteor
  • Drones, war, and reindustrialisation as national security
  • Humanoid vs dedicated robots; safety; robots that feel human (Pixar/Disney cues)
  • AI’s current (limited) role in CAD and hardware engineering; world models; the data problem
  • Building hardware orgs at Apple and Meta; the “back of the cabinet”; Quest 2 cost redesign
  • Hiring for zero-to-one: generalists, scalers, and “AI-native” new grads; mission + gut spark
  • Leaving OpenAI over the Department of War deal’s governance (“a third path”)
  • Lessons from Sam Altman (think 100×), Steve Jobs (unwavering bar), Mark Zuckerberg (push decisions down)

See also