Playbook
Tasteful Skills
“Tasteful Skills” argues that the best agent skills are not documentation or best-practice lists.
Back to Podcast Digest
Why we’re at the beginning of the AI hardware boom | Caitlin Kalinowski (ex–OpenAI, Meta, Apple)
TL;DR
Caitlin Kalinowski thinks AI is about to leave the keyboard and hit the physical world — after digital AI “saturates,” she says the next frontier is robotics, manufacturing, drones, autonomy, and eventually space, which is why hardware suddenly feels hot again at labs and startups.
VR didn’t fail so much as become the R&D lab for robotics — the money poured into Oculus, Quest, and AR work like Meta’s Orion produced core technologies such as SLAM, depth sensing, and spatial perception that now map directly onto robots, drones, and autonomous systems.
Hardware is brutally unforgiving compared with software — Kalinowski’s line is that hardware teams only get to “compile” four or five times total, not every day, so changing goals late, missing one component, or misreading a spec can trigger catastrophic redesigns and months of delay.
Humanoid robots are real, but still prototypes—and safety is the blocker people underrate — she points to 1X Neo’s softer, inward-mass design as the right instinct, and notes many strong robots today still come with warnings that no human should be within 3 feet.
The real bottleneck in AI hardware may be supply chain, not intelligence — magnets, actuators, batteries, RAM, and silicon all sit on fragile global dependencies, and she warns memory prices could double as AI data centers soak up supply, creating “a meteor” for robotics and consumer hardware.
The best builders she’s worked with pushed from different angles: Steve Jobs demanded excellence, Mark Zuckerberg optimized decision-making speed, and Sam Altman kept asking “why not 100x?” — those leadership styles shaped how she thinks about ambitious hardware teams, mission alignment, and zero-to-one hiring.
The Breakdown
Why VR never became mainstream—but still mattered
Lenny opens by asking the obvious question: after billions from Meta, Apple’s Vision Pro push, and years of effort, why didn’t VR really break through? Kalinowski’s answer is that VR taught the industry foundational spatial technologies—SLAM, depth sensing, orientation in 3D space—that now matter even more for robotics than for gaming. She says the social friction of putting something over your face was always a bigger issue than people wanted to admit.
AR glasses, Orion, and the path beyond phones
Kalinowski says she still believes in AR glasses because constantly looking down at phones is bad for how humans relate to each other. She describes Meta’s Orion glasses as “ahead of their time”: the waveguides and microLEDs worked, but yields and cost still make mass production hard. The promise, in her telling, is a mostly-off display that preserves social presence until you need information.
Why hardware is suddenly cool again
Lenny brings up Princeton students shifting from computer science into robotics, and Kalinowski says it feels genuinely strange because hardware was never the glamorous path. Her explanation for the swing is simple: AI progress is going “vertical,” and once software behind a keyboard starts to feel saturated, the next frontier is the physical world. That means robots, manufacturing, autonomy, sensing, and all the messy real-world systems software people used to ignore.
The brutal reality of building atoms instead of code
One of her sharpest metaphors lands here: software can compile constantly, but hardware only gets four or five “compiles” ever. Because products ship physically, teams have to solve tolerance stacks, reliability, yield, and variance before mass production, not after. She gives the unsexy truth of hardware: if one part is wrong—or one supplier disappears—you can trigger a board redesign, line retesting, and months of pain.
Humanoids, safety, and why soft robots matter
Asked about Tesla Optimus, Figure, and 1X Neo, Kalinowski says humanoids are still advanced prototypes, not finished products. The thing she keeps coming back to is safety: a strong robot moving near people is fundamentally an impact-risk problem, which is why softer, lighter, more compliant designs matter. She’s blunt that many current robots are still sold with rules saying humans shouldn’t come within 3 feet.
The supply chain problem underneath the robot boom
From there, the conversation turns geopolitical fast. Kalinowski walks through the stack from raw magnets to actuator motors to subassemblies, arguing that the U.S. outsourced too much of this foundation over 25 years to China, Japan, and Korea. Her takeaway is that reindustrialization isn’t nostalgic politics; it’s a military and economic necessity in a world where drones and robots depend on the same core components.
Drones, warfare, and the math we’re losing
This is the darkest and most memorable stretch of the episode. She agrees with Palmer Luckey’s view that in the next two years the U.S. should invest far more in drones than aircraft carriers, and points to Ukraine as the proof that military tech is now updating daily with AI and 3D printing. Her line that sticks is that there may be “more change in war than there is in consumer electronics” over the next two years.
How Apple, Meta, and OpenAI shaped her view of great teams
Kalinowski reflects on Apple as the place that taught a generation how to think through hardware trade-offs, risk, and craftsmanship—down to Steve Jobs’ “back of the cabinet” standard. At Meta, she saw a surprisingly crisp operating system for decision-making under Zuckerberg and Andrew Bosworth, and at OpenAI she says Sam Altman constantly pushed people to think 100x bigger. She also explains why she left OpenAI after the Department of Defense-related announcement: not scorched earth, but a clear boundary around governance and how that decision was made.
What AI can and can’t do for hardware yet
She’s excited about AI in engineering, but very specific about what’s missing. Today’s tools help with planning, research, spreadsheets, PCB routing, and maybe rough geometry, but they still don’t truly understand friction, contact, mass, or folding a piece of paper in physical space. Her ask is memorable: she wants the equivalent of Codex for hardware engineering—and thinks getting there may require new world-model-style systems plus access to proprietary CAD data that companies are reluctant to share.
Robots people actually want around them
In the closing stretch, Kalinowski gets into what makes robots feel human rather than creepy, drawing on roboticist Leila Takayama’s research. A robot should acknowledge you, signal intent before moving, and seem soft, aware, and non-threatening; otherwise it triggers the uncanny “it’s just staring” reaction. She says Pixar and Disney are probably the best in the world at this kind of emotional and behavioral design, which is a very human note to end a deeply technical conversation.
Share
Keep Reading
Two Ways to Follow
The Weekly Echo. The inbox-shaped summary of what mattered.
New editorials announced here.
